This paper proposed an end-to-end road crack segmentation model based on attention mechanism and deep FCN with generative adversarial learning. We create a segmentation network by introducing a visual attention mechanism and residual module to a fully convolutional network(FCN) to capture richer local features and more global semantic features and get a better segment result. Besides, we use an adversarial network consisting of convolutional layers as a discrimination network. The main contributions of this work are as follows: 1) We introduce a CNN model as a discriminate network to realize adversarial learning to guide the training of the segmentation network, which is trained in a min-max way: the discrimination network is trained by maximizing the loss function, while the segmentation network is trained with the only gradient passed by the discrimination network and aim at minimizing the loss function, and finally an optimal segmentation network is obtained; 2) We add the residual modular and the visual attention mechanism to U-Net, which makes the segmentation results more robust, refined and smooth; 3) Extensive experiments are conducted on three public road crack datasets to evaluate the performance of our proposed model. Qualitative and quantitative comparisons between the proposed method and the state-of-the-art methods show that the proposed method outperforms or is comparable to the state-of-the-art methods in both F1 score and precision. In particular, compared with U-Net, the mIoU of our proposed method is increased about 3%~17% compared with the three public datasets.
Citation: Xing Hu, Minghui Yao, Dawei Zhang. Road crack segmentation using an attention residual U-Net with generative adversarial learning[J]. Mathematical Biosciences and Engineering, 2021, 18(6): 9669-9684. doi: 10.3934/mbe.2021473
| [1] | Chun Nok Lam, William Nicholas, Alejandro De La Torre, Yanpui Chan, Jennifer B. Unger, Neeraj Sood, Howard Hu . Factors associated with parents' willingness to vaccinate their children against COVID-19: The LA pandemic surveillance cohort study. AIMS Public Health, 2022, 9(3): 482-489. doi: 10.3934/publichealth.2022033 |
| [2] | Kassidy C Colton, Stephanie A Godleski, Joseph S Baschnagel, Rebecca J Houston, Shine M DeHarder . Alcohol use during the COVID-19 pandemic: gender, parenthood, intimate partner violence, and stress. AIMS Public Health, 2023, 10(2): 360-377. doi: 10.3934/publichealth.2023027 |
| [3] | Ali Moussaoui, El Hadi Zerga . Transmission dynamics of COVID-19 in Algeria: The impact of physical distancing and face masks. AIMS Public Health, 2020, 7(4): 816-827. doi: 10.3934/publichealth.2020063 |
| [4] | Shruthi Venkatesh, Alexandra Gill, Lauren Kim, Stacey N Doan . Mothers' intentions to vaccinate their children for COVID-19. AIMS Public Health, 2023, 10(1): 209-218. doi: 10.3934/publichealth.2023016 |
| [5] | Clemens Drenowatz, Gerson Ferrari, Klaus Greier, Sitong Chen, Franz Hinterkörner . Physical fitness in Austrian elementary school children prior to and post-COVID-19. AIMS Public Health, 2023, 10(2): 480-495. doi: 10.3934/publichealth.2023034 |
| [6] | Sophie Terp, Sameer Ahmed, Elizabeth Burner, Madeline Ross, Molly Grassini, Briah Fischer, Parveen Parmar . Deaths in Immigration and Customs Enforcement (ICE) detention: FY2018–2020. AIMS Public Health, 2021, 8(1): 81-89. doi: 10.3934/publichealth.2021006 |
| [7] | Sameer Badri Al-Mhanna, Alexios Batrakoulis, Abdulrahman M. Sheikh, Abdulaziz A. Aldayel, Abdulwali Sabo, Mahaneem Mohamed, Hafeez Abiola Afolabi, Abdirizak Yusuf Ahmed, Sahra Isse Mohamed, Mehmet Gülü, Wan Syaheedah Wan Ghazali . Impact of COVID-19 lockdown on physical activity behavior among students in Somalia. AIMS Public Health, 2024, 11(2): 459-476. doi: 10.3934/publichealth.2024023 |
| [8] | Christos Sikaras, Ioannis Ilias, Athanasios Tselebis, Argyro Pachi, Sofia Zyga, Maria Tsironi, Andrea Paola Rojas Gil, Aspasia Panagiotou . Nursing staff fatigue and burnout during the COVID-19 pandemic in Greece. AIMS Public Health, 2022, 9(1): 94-105. doi: 10.3934/publichealth.2022008 |
| [9] | Chloe Zivot, Cate Dewey, Meghan Brockington, Chioma Nwebube, Ghaid Asfour, Natasha Vattikonda, Debbie Bell, Sharada Srinivasan, Matthew Little . Experiences of wellbeing and resilience among refugee mothers and families in Calgary during the COVID-19 pandemic, and the role of participation in HIPPY, a home visiting program. AIMS Public Health, 2022, 9(3): 521-541. doi: 10.3934/publichealth.2022036 |
| [10] | Klaus Greier, Clemens Drenowatz, Theresa Bischofer, Gloria Petrasch, Carla Greier, Armando Cocca, Gerhard Ruedl . Physical activity and sitting time prior to and during COVID-19 lockdown in Austrian high-school students. AIMS Public Health, 2021, 8(3): 531-540. doi: 10.3934/publichealth.2021043 |
This paper proposed an end-to-end road crack segmentation model based on attention mechanism and deep FCN with generative adversarial learning. We create a segmentation network by introducing a visual attention mechanism and residual module to a fully convolutional network(FCN) to capture richer local features and more global semantic features and get a better segment result. Besides, we use an adversarial network consisting of convolutional layers as a discrimination network. The main contributions of this work are as follows: 1) We introduce a CNN model as a discriminate network to realize adversarial learning to guide the training of the segmentation network, which is trained in a min-max way: the discrimination network is trained by maximizing the loss function, while the segmentation network is trained with the only gradient passed by the discrimination network and aim at minimizing the loss function, and finally an optimal segmentation network is obtained; 2) We add the residual modular and the visual attention mechanism to U-Net, which makes the segmentation results more robust, refined and smooth; 3) Extensive experiments are conducted on three public road crack datasets to evaluate the performance of our proposed model. Qualitative and quantitative comparisons between the proposed method and the state-of-the-art methods show that the proposed method outperforms or is comparable to the state-of-the-art methods in both F1 score and precision. In particular, compared with U-Net, the mIoU of our proposed method is increased about 3%~17% compared with the three public datasets.
In Europe, eco-driving has become a popular measure for coping with global warming issues in the transport sector. For example, “ecodrive.org” listed five “Golden Rules of Eco-driving”: 1. Anticipate traffic flow; 2. Maintain a steady speed at low RPM; 3. Shift up early; 4. Check tire pressures frequently at least once a month and before driving at high speed; and 5. Consider any extra energy required costs fuel and money [1]. In Japan, “10 Recommendations for Eco-driving” are listed in [2] and [3]. They are mostly consistent with those listed above except for “Press the accelerator gently when accelerating, ” which is strongly recommended. The correspondence between eco-driving tips in Europe and in Japan is shown in Table 1. Differences in the lists arose from differences in the predominant transmission types of passenger vehicles. Manual transmissions are mainstream in Europe, while automatic transmissions are preferred in Japan. The instructions to “shift up early at low RPM” (in Europe) and “go easy on the acceleration pedal” (in Japan) are both intended to improve the engine’s energy-conversion efficiency. There are many reports on the effectiveness of eco-driving as a results of drivers’ spontaneous behavior change; ecodrive.org showed that eco-driving saves 5%-15% of fuel-consumption over the long term [1]. Lovejoy et al. [4] reviewed previous studies in the U.S. concerning eco-driving and reported that it reduced fuel consumption by 10%-31%. Kato and Kobayashi [5] reviewed many reports [5,6,7] in Japan and reported a saving of 10%-20% from eco-driving. Kato and Kobayashi [5] also reported that eco-driving in a test-ride event reduced fuel consumption by 11.6%, and its major effect was to decrease the kinetic running energy due to observation of the speed limit andthe maintenance of a constant speed. EV users’ interest in eco-driving stems from maximizing the travel distance per charge of their EVs [8]. However, discussions and quantitative evaluations of whether eco-driving methods for ICEVs are valid for EVs are rare. Therefore, this study conducted comparative measurements of the effects of eco-driving for ICEVs and EVs using a chassis dynamometer (C/D), which could provide a constant test environment in a room at a laboratory.
| “five Golden Rules of Eco-driving” in Europe | “10 Recommendations for Eco-driving” in Japan |
| 1. Anticipate traffic flow | - Release the accelerator earlier when decelerating |
| 2. Maintain a steady speed at low RPM | - Reduce acceleration and deceleration while keeping enough distance between cars |
| 3. Shift up early | - |
| 4. Check tire pressures frequently at least once a month and before driving at high speed | - Check the pressure of the tires as the first step toward better maintenance |
| 5. Consider any extra energy required costs fuel and money |
- Use air conditioners appropriately - Avoid unnecessary idling - Avoid traffic jams; leave home with time to spare - Take out unnecessary loads - Be aware of your fuel consumption |
| - |
- Press the accelerator gently when accelerating - Do not block traffic when parking |
DownLoad: CSVThe method for evaluating reduction of fuel consumption and CO2 emissions through eco-driving which the same subjects drive twice (once as usual and once eco-driving) along a test route on a public road is conventionally and mainly used. Even when comparing several vehicle models, it is not impossible to verify the difference in effect between vehicles or to perform the actual running test on public roads. However, it may be difficult to evaluate the eco-driving effect because the subjects might not be the same, conditions such as weather, road conditions are not kept constant and these factors may also affect fuel consumption. In this study, operation with the intention of eco-driving for a variety of vehicles was evaluated using a C/D facility capable of performing the test under constant conditions. The test method had three steps, as shown in Figure 1. Firstly, we created an eco-driving test mode from the results if a real running test that was conducted using a gasoline-powered passenger car. Then, these tests were performed for a plurality of models, including gasoline and electric vehicles, using a C/D facility.
Figure 1. Flow of the test methodTo obtain the data concerning eco-driving achievement level, an eco-driving test-ride event was conducted [5]. Twenty six test subjects drove vehicles along a test route which was mainly on a public road in Tsukuba-city, Japan. The subjects were thirteen females and thirteen males who were non-professional drivers working for the National Institute for Environmental Studies (NIES) in Japan. Their ages were between 20 and 65 years old. They mainly commuted by car. The test route was 5.2 km long, comprising a 3.7-km arterial road with a speed limit of 60km/h, a 0.9-km community road with a speed limit of 40km/h, and a 0.6-km private road in the NIES with a speed limit of 20 km/h. Speed exceeding the limit by 10 km/h or more were commonly observed on the arterial road. The test route imitated typical commuter routes in Japanese local cities. The vehicle used in this event was Japan’s best-selling passenger car, which is equipped with a gasoline-powered, 1,300 cc displacement engine and a continuously variable transmission (CVT). The equipment used to collect data from driving was an “eco-manager” developed by the Environmental Restoration and Conservation Agency (ERCA) in Japan. It was linked to the vehicle’s engine control unit (ECU) and logged data concerning vehicle speed, fuel consumption, and so on with a 1-second collection interval. Each subject drove the passenger car twice; once as normal, and then immediately after in eco-driving. Before the second trip, an instructor gave subjects guidance pertaining to the three eco-driving tips listed in Table 2. The only tips related to driving operation were selected from 10 eco-driving tips. However, the manual idling stop while waiting at signals was prohibited because of safety considerations. Furthermore, the use of air conditioner was also prohibited in order to unify the accessory-usage conditions.
|
1. Observe the speed limit and maintain a constant speed 2. Release the accelerator earlier when decelerating 3. Press the accelerator gently when accelerating |
DownLoad: CSVThe average fuel consumption rates during normal driving and during eco-driving were 60.4 and 53.4 cc/km, respectively. As a result of a paired t-test, it was confirmed that fuel consumption during eco-driving was less than that during normal driving at a significance level of 99%. The average fuel and CO2-reduction rates were 11.6%. Gasoline consumption [cc/km] and CO2 emissions [g/km] were proportional since only the CO2 emissions of the tank-to-wheel were considered in this study. Therefore, the reduction rates of fuel consumption and CO2 emissions were calculated using formula (1). We call the reduction rates the eco-driving effects:
| RR=FC0−FC1FC0×100 | (1) |
where RR is the reduction rate of fuel consumption and CO2 emissions [%], FC0 is fuel consumption during normal driving [cc/km], and FC1 is fuel consumption during eco-driving [cc/km].
To objectively express the eco-driving achievement level, the kinetic running energy [J] was used as an indicator. This energy was calculated using formulas (2) through (9) (Kato et al. [9]):
| E=T∑t=0,P>0P(t)⋅Δt1000×D; | (2) |
| P(t)=R(t)⋅V(t); | (3) |
| R(t)=Rr+Rl(t)+Ra(t); | (4) |
| Rr=μ⋅M⋅g; | (5) |
| Rl(t)=ρ⋅Cd⋅S⋅V(t)2/2; | (6) |
| Ra(t)=(M+δ⋅M)⋅α(t); | (7) |
| D=T∑t=0V(t)1000; | (8) |
| α(t)=V(t)−V(t−Δt)Δt; | (9) |
where T is the travel time [sec], D is the travel distance [km], P(t) is the power required to propel the vehicle [W], R(t) is the running resistance [N], V(t) is the vehicle speed [m/s], Rr is the rolling frictional resistance, Rl(t) is the air resistance, Ra(t) is the acceleration resistance, μ is the coefficient of friction [-], M is the vehicle mass including two passengers [kg], g is the gravitational constant [9.8 m/s2], ρ is the air density at standard ambient temperature and pressure [1.169 kg/m3], Cd is the coefficient of air resistance -, S is the frontal projected area [m2], δM is the equivalent mass of the rotating parts [kg] and a(t) is the acceleration [m/s2]. The calculation step, Dt, is one second for all formulas except for formulas (5). The vehicle speed data obtained by the “eco-manager” was used as V(t). The vehicle specification used for the calculation are shown in Table 3. Due to lack of road-inclination data, gradient resistance is not taken into account.
| parameter | value |
| μ - | 0.013 |
| M [kg] | 1,110 |
| Cd - | 0.35 |
| S [m2] | 2.16 |
| δM [kg] | 38 |
DownLoad: CSVFigure 2 shows the relationship between kinetic running energy and fuel-consumption rate for all fifty-two driving datasets. The orange and green points correspond to normal driving and eco-driving, respectively. A good correlation was found between the two quantities. By adopting eco-driving, kinetic running energy and fuel consumption were on average reduced by 15.5% and 11.6%, respectively.
Figure 2. Relationship between kinetic running energy and fuel-consumption rate in the eco-driving test-ride event (real-world driving data)To express the relationship between the kinetic running energy and fuel (or electric) consumption rate, four speed patterns were selected from the fifty-two collected in the eco-driving test-ride event (the four red circles in Figure 2). To determine suitable speed patterns, characteristics such as the equality of the running energy’s interval and the similarity of stopping frequencies and intersections at which stopping took place were taken into account. Most trips had 5-6 stops. Trips were classified into groups based on the intersections at which they stopped. The most appropriate trip group was selected based on the variation of the kinetic running energy and by matching the regression line. We called the set of four speed patterns the “eco-driving test mode, ” and named the speed patterns “ECO-S, ” “ECO-A, ” “ECO-B, ” and “ECO-C” in increasing order of kinetic running energy. ”ECO-S” has the lowest kinetic running energy. Table 4 shows the specifications of each eco-driving test mode. The speed patterns of these modes are shown in Figure 3.
| ECO-S | ECO-A | ECO-B | ECO-C | ||
| Travel Time (sec) | 648 | 628 | 627 | 612 | |
| Max Speed (km/h) | 56 | 64 | 66 | 80 | |
| Stopping Frequency | 5 | 5 | 6 | 5 | |
| Running Energy (kJ/km) | 277 | 316 | 395 | 463 | |
| Fuel Consumption (cc/km) | 45.3 | 48.5 | 59.7 | 67.5 | |
| Time Share | Idle | 14% | 20% | 26% | 28% |
| Cruise | 40% | 35% | 29% | 20% | |
| Acceleration | 23% | 24% | 26% | 24% | |
| Deceleration | 23% | 20% | 18% | 29% | |
DownLoad: CSV
Figure 3. Four speed patterns constituting the “eco-driving test mode.”It is not practical to perform running tests targeting a variety of vehicles in the real world due to time constraints, economic constraints, difficulties maintaining similar environmental conditions and subject attributes, and so on. A C/D can test a variety of vehicles under the same conditions. In this study, three ICEVs, one HEV, and two EVs were tested using a C/D with an eco-driving test mode. Table 5 shows the specifications of each vehicle. The 660 cc ICEV and both EVs were small passenger cars categorized as “kei-cars”—a Japanese unique super-minivehicle segment. A kei-car is less than or equal to 3.40 m in length, 1.48 m in width and 2.0 m in height. Their engine displacement is 660 cc or less, and the self-regulated maximum power of the engine or motor is 47 kW or less. About one-third of car ownership in Japan is of kei-cars. One EV had front-wheel drive (called “EV type A” in this paper), and the other had rear-wheel drive (called “EV type B”). EV type A is not commercially available. Some views of the C/D test are shown in Figure 4.
| Vehicle Type | Engine Displacement | Transmission | Weight (nt.) | |
| 660cc CVT | ICEV | 1,800 cc | CVT | 770 kg |
| 1,300cc CVT | 1,300 cc | CVT | 1,000 kg | |
| 1,800cc 4AT | 660 cc | 4 AT | 1,170 kg | |
| 1,500cc HEV | HEV | 1,500 cc | e-CVT | 1,260 kg |
| EV Type A | EV | - | - | 1,030 kg |
| EV Type B | - | - | 1,070 kg |
DownLoad: CSV
Figure 4. View of chassis-dynamometer testIn order to confirm its reproducibility, the eco-driving effect in the test-ride event was compared to the results of the C/D test for the same vehicle, the 1,300 cc CVT. Figure 5 shows the relevant test results. The four circular markers in this figure represent the fuel-consumption rates of ECO-S, ECO-A, ECO-B and ECO-C. A good linear relationship between the kinetic running energy and fuel-consumption rate was found. Using this linear-regression equation, the eco-driving effect was estimated. The origin and end point of the arrow in this figure were the average kinetic running energy for “normal driving” and “eco-driving, ” respectively, over all twenty-six subjects. The fuel consumption rate was decreased from 51.5 to 45.2 cc/km. Therefore, the eco-driving effect estimated by the C/D test was 12.2%. Considering that the eco-driving effect in the test-ride event was 11.6%, it is considered to be highly reproducible with an error of less than 5%.
Figure 5. Test results of 1300 cc CVTThe eco-driving effects of the five other vehicles besides the 1,300 cc CVT were also estimated using the same way as previously mentioned at 3.1. Confirmation of Reproducibility. Figure 6 shows the estimated eco-driving effects of all six vehicles. The eco-driving effects of 660 cc CTV, 1,300 cc CVT, 1,800 cc 4AT, 1,500 cc HEV, EV type A, and EV type B were 12.0%, 12.2%, 10.9%, 12.6%, 18.4% and 11.7%, respectively. The results indicate that the eco-driving tips for ICEVs were effective not only for ICEVs (including HEVs), but also for EVs.
Figure 6. Estimated eco-driving effectEV type A had a larger eco-driving effect than the other vehicles. To determine the reason for this, the energy-conversion efficiency, H [%], was calculated using formulas (10) and (11):
For ICEVs and a HEV,
| H=EFC×34.6×100 | (10) |
For EVs,
| H=EEC×3,600×100 | (11) |
where E is the kinetic running energy [kJ/km], FC is the fuel-consumption rate [cc/km], and EC is the electric-consumption rate [kWh/km].
Figure 7 shows the test results for the energy-conversion efficiencies of all vehicles. From the viewpoint of differences in energy efficiency due to the type of vehicle, the averaged efficiency of 660 cc CTV, 1,300 cc CVT, 1,800 cc 4AT, 1,500 cc HEV, EV type A, and EV type B were 16.8%, 21.3%, 16.7%, 30.7%, 83.5%, and 80.8% respectively. It can be concluded that the hybrid system increased the utilization efficiency of thermal energy by 9.4-14.0%, and it is well known that the motors of EVs can convert electrical energy into kinetic energy with high efficiency. On the other hand, from the viewpoint of the differences in energy efficiency due to eco-driving achievement level, the energy-conversion efficiency tended to decrease with low kinetic running energy, except for the EV type A. This vehicle maintained high energy efficiency over a wide range of kinetic running energies. This result indicates that EVs have higher potential eco-driving effects than ICEVs if they maintain high energy-conversion efficiencies at low load range, because ICEV engines have a characteristically low energy-conversion efficiency at low load range. We believe that the EV type B had a lower energy efficiency in this condition because the regenerative energy during deceleration of a rear-wheel-drive vehicle is less than that of a front-wheel-drive vehicle.
Figure 7. Test results of all vehicles with respect to energy conversion efficiencyComparative measurements of the eco-driving effect between ICEVs and EVs were conducted using a C/D with an eco-driving test mode that we developed. Eco-driving effects and energy-conversion efficiencies were examined. The eco-driving effects of 660 cc CVT, 1,300 cc CVT, 1,800 cc 4AT, 1,500 cc HEV, EV type A, and EV type B vehicles were 12.0%, 12.2%, 10.9%, 12.6%, 18.4% and 11.7% respectively. These results indicate that eco-driving with low kinetic running energy by “observing the speed limit” and “maintaining a constant speed” is effective not only for ICEVs (including HEVs), but also for EVs. While, the averaged energy-conversion efficiency of 660 cc CTV, 1,300 cc CVT, 1,800 cc 4AT, 1,500 cc HEV, EV type A, and EV type B were 16.8%, 21.3%, 16.7%, 30.7%, 83.5%, and 80.8% respectively. Electric vehicles have high energy-conversion efficiencies because they use a motor to convert electrical energy into kinetic energy. From the viewpoint of the differences in energy efficiency due to eco-driving achievement level, the energy-conversion efficiency of tested vehicles tended to decrease with low kinetic running energy, except for the EV type A which had the largest eco-driving effect. These results indicate that EVs have a higher potential of eco-driving effect than a gasoline-powered vehicle if EVs could maintain high energy-conversion efficiency at low load range. EVs have begun to be more widely adopted, and various power trains will coexist on public roads. It is preferable for the same eco-driving method to be used by both electric and gasoline-powered vehicles to create a smooth and safe traffic flow. At present, it remains the responsibility of the driver to observe the speed limit and maintain a constant speed. Therefore, this study is intended to support the dissemination of tools like intelligent speed adaptation (ISA) to obey speedregulations in real time. In the future, as well as the development and dissemination of automated driving systems, traveling with less kinetic energy will be important. The future steps of this study are as follows: it is necessary to extend the evaluation target of electric vehicles. The eco-driving test mode developed in this study assumed a Japanese local city with an upgraded arterial road. If the road conditions are different (frequent road congestion, etc.), it will be necessary to create a new test mode in order to evaluate them.
We would like to express our sincere appreciation to everyone who attend the eco-driving test-ride event. This work was supported by JSPS KAKENHI Grant Number 24310116.
All authors declare no conflict of interest in this paper.
| [1] | Li. Q, Liu. X, Novel approach to pavement image segmentation based on neighboring difference histogram method, in 2008 Congress on Image and Signal Processing, IEEE, (2008), 792–796. |
| [2] | M. S. Kaseko, S. G. Ritchie, A neural network-based methodology for pavement crack detection and classification, Transport. Res. C.-Emer., 1 (1993), 275–291. |
| [3] | M. Gavilán, D. Balcones, O. Marcos, Adaptive road crack detection system by pavement classification, Sensors, 11 (2011), 9628–9657. |
| [4] | T. S. Nguyen, S. Begot, F. Duculty, Free-form anisotropy: A new method for crack detection on pavement surface images, in 2011 18th IEEE International Conference on Image Processing, IEEE, (2011), 1069–1072. |
| [5] | R. Amhaz, S. Chambon, J. Idier, Automatic crack detection on two-dimensional pavement images: An algorithm based on minimal path selection, IEEE. T. Intell. Transp., 17 (2016), 2718–2729. |
| [6] | M. Avila, S. Begot, F. Duculty, 2D image based road pavement crack detection by calculating minimal paths and dynamic programming, in 2014 IEEE International Conference on Image Processing (ICIP), IEEE, (2014), 783–787. |
| [7] | Q. Li., D. Zhang, Q. Zou, 3D laser imaging and sparse points grouping for pavement crack detection, In: A. Scarpas, N. Kringos, I. Al-Qadi, Loizos A, eds, in 2017 25th European Signal Processing Conference (EUSIPCO), (2017), 2036–2040. |
| [8] | Q. Zou, Y. Cao, Q. Li, CrackTree: Automatic crack detection from pavement images, Pattern. Recogn. Lett., 33 (2012), 227–238. |
| [9] | Y. Huang, Y. J. Tsai, Crack fundamental element (CFE) for multi-scale crack classification, in 7th RILEM International Conference on Cracking in Pavements, (2012), 419–428. |
| [10] | Y. J. Tsai, C. Jiang, Z. Wang, Implementation of automatic crack evaluation using crack fundamental element, in 2014 IEEE International Conference on Image Processing (ICIP), IEEE, (2014), 773–777. |
| [11] | Y. Chen, Y. Zhang, J. Yang, Curve-like structure extraction using minimal path propagation with backtracking, IEEE T. Image. Process, 25 (2015), 988–1003. |
| [12] | K. Y. Song, M. Petrou, J. Kittler, Texture crack detection, Mach. Vision. Appl, 8 (1995): 63–75. |
| [13] | M. Petrou, J. Kittler, K. Y. Song, Automatic surface crack detection on textured materials, J. Mater. Process. Tech., 56 (1996), 158–167. |
| [14] | E. Douka, S. Loutridis, A. Trochidis, Crack identification in plates using wavelet analysis, J. Sound. Vib., 270 (2004), 279–295. |
| [15] | P. Subirats, J. Dumoulin, V. Legeay, Automation of pavement surface crack detection using the continuous wavelet transform, in 2006 International Conference on Image Processing, IEEE, (2006), 3037–3040. |
| [16] | L. Itti, C. Koch, E. Niebur, A model of saliency-based visual attention for rapid scene analysis, IEEE T. Pattern. Anal., 20 (1998), 1254–1259. |
| [17] | W. Xu, Z. Tang, J. Zhou, Pavement crack detection based on saliency and statistical features, in 2013 IEEE International Conference on Image Processing, IEEE, (2013), 4093–4097. |
| [18] | D. Ciresan, A. Giusti, L. Gambardella, Deep neural networks segment neuronal membranes in electron microscopy images, Adv. Neural Inform. Process Syst., 25 (2012), 2843–2851. |
| [19] | Y. J. Cha, W. Choi, O. Büyüköztürk, Deep learning-based crack damage detection using convolutional neural networks, Computer‐Aided Civil Infrast. Eng., 32 (2017), 361–378. |
| [20] | Y. J. Cha, W. Choi, G. Suh, Autonomous structural visual inspection using region-based deep learning for detecting multiple damage types, Computer‐Aided Civil Infrast. Eng., 33 (2018), 731–747. |
| [21] | Y. Liu, J. Yao, X. Lu, DeepCrack: A deep hierarchical feature learning architecture for crack segmentation, Neurocomputing, 338 (2019), 139–153. |
| [22] | J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in Proceedings of the IEEE conference on computer vision and pattern recognition, IEEE, (2015), 3431–3440. |
| [23] | M. M. Islam, J. M. Kim, Vision-based autonomous crack detection of concrete structures using a fully convolutional encoder–decoder network, Sensors, 19 (2019), 4251. |
| [24] | O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, Springer International Publishing, 2015. |
| [25] | O. Oktay, J. Schlemper, L. L. Folgoc, Attention u-net: Learning where to look for the pancreas, 2018. |
| [26] | Z. Liu, Y. Cao, Y. Wang, Computer vision-based concrete crack detection using U-net fully convolutional networks, Automat. Constr., 104 (2019), 129–139. |
| [27] | V. Badrinarayanan, A. Kendall, R. Cipolla, Segnet: A deep convolutional encoder-decoder architecture for image segmentation, IEEE T. Pattern. Anal., 39 (2017), 2481–2495. |
| [28] | Q. Zou, Z. Zhang, Q. Li, Deepcrack: Learning hierarchical convolutional features for crack detection, IEEE T. Image. Process, 28 (2018), 1498–1512. |
| [29] | Y. Liu, J. Yao, X. Lu, DeepCrack: A deep hierarchical feature learning architecture for crack segmentation, Neurocomputing, 338 (2019), 139–153. |
| [30] | F. Wang, M. Jiang, C. Qian, Residual attention network for image classification, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, (2017), 3156–3164. |
| [31] | D. Yang, H. R. Karimi, K. Sun, Residual wide-kernel deep convolutional auto-encoder for intelligent rotating machinery fault diagnosis with limited samples, Neural Networks, 141 (2021), 133–144. |
| [32] | J. Huyan, W. Li, S. Tighe, CrackU‐net: A novel deep convolutional neural network for pixelwise pavement crack detection, Struct. Contro. Hlth., 27 (2020), e2551. |
| [33] | Z. Fan, C. Li, Y. Chen, Ensemble of deep convolutional neural networks for automatic pavement crack detection and measurement, Coatings, 10 (2020), 152. |
| [34] | W. Song, G. Jia, D. Jia, Automatic pavement crack detection and classification using multiscale feature attention network, IEEE Access, 7 (2019), 171001–171012. |
| [35] | I. Goodfellow, J. Pouget-Abadie, M. Mirza, Generative adversarial networks, Commun. ACM, 63 (2020), 139–144. |
| [36] | P. Luc, C. Couprie, S. Chintala, Semantic segmentation using adversarial networks, in NIPS Workshop on Adversarial Training, 2016. |
| [37] | N. Souly, C. Spampinato, M. Shah, Semi supervised semantic segmentation using generative adversarial network, in Proceedings of the IEEE conference on computer vision and pattern recognition, IEEE, (2017), 5688–5696. |
| [38] | G. Wu, Q. Wang, D. Zhang, A generative probability model of joint label fusion for multi-atlas based brain segmentation, Med. Image. Anal., 18 (2014), 881–890. |
| [39] | V. Alex, M. S. KP, S. S. Chennamsetty, Generative adversarial networks for brain lesion detection, in Medical Imaging 2017: Image Processing-International Society for Optics and Photonics, (2017), 10133: 101330G. |
| [40] | Z. Gao, B. Peng, T. Li, Generative adversarial networks for road crack image segmentation, in 2019 International Joint Conference on Neural Networks (IJCNN), IEEE, (2019), 1–8. |
| [41] | D. Bahdanau, K. Cho, Y. Bengio, Neural machine translation by jointly learning to align and translate, in 3rd International Conference on Learning Representations(ICLR), 2015. |
| [42] | I. J. Goodfellow, J. Shlens, C. Szegedy, Explaining and harnessing adversarial examples, 2014. |
| [43] | A. Madry, A. Makelov, L. Schmidt, Towards deep learning models resistant to adversarial attacks, in International Conference on Learning Representations, 2018. |
| [44] | Y. Shi, L. Cui, Z. Qi, Automatic road crack detection using random structured forests, IEEE T. Intell. Transp., 17 (2016), 3434–3445. |
| [45] | N. T. H. Nguyen, T. H. Le, S. Perry, Pavement crack detection using convolutional neural network, in Proceedings of the Ninth International Symposium on Information and Communication Technology, (2018), 251–256. |
| [46] | M. D. Jenkins, T. A. Carr, M. I. Iglesias, A deep convolutional neural network for semantic pixel-wise segmentation of road and pavement surface cracks, in 2018 26th European Signal Processing Conference (EUSIPCO), IEEE, (2018), 2120–2124. |
| [47] | X. Weng, Y. Huang, W. Wang, Segment-based pavement crack quantification, Automat. Constr., 105 (2019), 102819. |
| 1. | George Goes, Renata Bandeira, Daniel Gonçalves, Márcio de Almeida D'Agosto, Cíntia Oliveira, The effect of eco-driving initiatives toward sustainable urban waste collection, 2020, 14, 1556-8318, 569, 10.1080/15568318.2019.1584933 | |
| 2. | Giuseppe Mauromicale, Angelo Raciti, Santi Agatino Rizzo, Giovanni Susinni, Luigi Abbatelli, Simone Buonomo, Vittorio Giuffrida, Alessandra Raffa, 2019, Improvement of SiC power module layout to mitigate the gate-source overvoltage during switching operation, 978-8-8872-3743-6, 1, 10.23919/EETA.2019.8804487 | |
| 3. | Luigi Abbatelli, Angelo Raciti, Rosario Scollo, Giuseppe Mauromicale, Santi Agatino Rizzo, Alfio Scuto, Domenico Nardo, Nunzio Salerno, Giovanni Susinni, 2019, Effects of parasitic phenomena in half bridge with Super Junction MOSFETs suitable for UAV, 978-8-8872-3745-0, 1, 10.23919/AEIT.2019.8893414 | |
| 4. | Liu Pai, Tomonobu Senjyu, A Yearly Based Multiobjective Park-and-Ride Control Approach Simulation Using Photovoltaic and Battery Energy Storage Systems: Fuxin, China Case Study, 2022, 14, 2071-1050, 8655, 10.3390/su14148655 | |
| 5. | Yasutaka Okada, Shunki Nishii, Akihiro Takeshita, Kazuki Harada, Yudai Yamasaki, Torque control strategy with V2X information for HEVs to minimize fuel consumption, 2022, 20, 2095-6983, 197, 10.1007/s11768-022-00094-y | |
| 6. | Xia Jiang, Jian Zhang, Xiaoyu Shi, Jian Cheng, Learning the Policy for Mixed Electric Platoon Control of Automated and Human-Driven Vehicles at Signalized Intersection: A Random Search Approach, 2023, 24, 1524-9050, 5131, 10.1109/TITS.2023.3242678 | |
| 7. | Saba Sabet, Bilal Farooq, Exploring the combined effects of major fuel technologies, eco-routing, and eco-driving for sustainable traffic decarbonization in downtown Toronto, 2025, 192, 09658564, 104385, 10.1016/j.tra.2025.104385 | |
| 8. | Anbazhagan Geetha, S. Usha, J. Santhakumar, Surender Reddy Salkuti, Forecasting of energy consumption rate and battery stress under real-world traffic conditions using ANN model with different learning algorithms, 2025, 13, 2333-8334, 125, 10.3934/energy.2025005 |
Figures(8) / Tables(3)
Xing Hu, Minghui Yao, Dawei Zhang. Road crack segmentation using an attention residual U-Net with generative adversarial learning[J]. Mathematical Biosciences and Engineering, 2021, 18(6): 9669-9684. doi: 10.3934/mbe.2021473
| “five Golden Rules of Eco-driving” in Europe | “10 Recommendations for Eco-driving” in Japan |
| 1. Anticipate traffic flow | - Release the accelerator earlier when decelerating |
| 2. Maintain a steady speed at low RPM | - Reduce acceleration and deceleration while keeping enough distance between cars |
| 3. Shift up early | - |
| 4. Check tire pressures frequently at least once a month and before driving at high speed | - Check the pressure of the tires as the first step toward better maintenance |
| 5. Consider any extra energy required costs fuel and money |
- Use air conditioners appropriately - Avoid unnecessary idling - Avoid traffic jams; leave home with time to spare - Take out unnecessary loads - Be aware of your fuel consumption |
| - |
- Press the accelerator gently when accelerating - Do not block traffic when parking |
DownLoad: CSV|
1. Observe the speed limit and maintain a constant speed 2. Release the accelerator earlier when decelerating 3. Press the accelerator gently when accelerating |
DownLoad: CSV| parameter | value |
| μ - | 0.013 |
| M [kg] | 1,110 |
| Cd - | 0.35 |
| S [m2] | 2.16 |
| δM [kg] | 38 |
DownLoad: CSV| ECO-S | ECO-A | ECO-B | ECO-C | ||
| Travel Time (sec) | 648 | 628 | 627 | 612 | |
| Max Speed (km/h) | 56 | 64 | 66 | 80 | |
| Stopping Frequency | 5 | 5 | 6 | 5 | |
| Running Energy (kJ/km) | 277 | 316 | 395 | 463 | |
| Fuel Consumption (cc/km) | 45.3 | 48.5 | 59.7 | 67.5 | |
| Time Share | Idle | 14% | 20% | 26% | 28% |
| Cruise | 40% | 35% | 29% | 20% | |
| Acceleration | 23% | 24% | 26% | 24% | |
| Deceleration | 23% | 20% | 18% | 29% | |
DownLoad: CSV| Vehicle Type | Engine Displacement | Transmission | Weight (nt.) | |
| 660cc CVT | ICEV | 1,800 cc | CVT | 770 kg |
| 1,300cc CVT | 1,300 cc | CVT | 1,000 kg | |
| 1,800cc 4AT | 660 cc | 4 AT | 1,170 kg | |
| 1,500cc HEV | HEV | 1,500 cc | e-CVT | 1,260 kg |
| EV Type A | EV | - | - | 1,030 kg |
| EV Type B | - | - | 1,070 kg |
DownLoad: CSV| “five Golden Rules of Eco-driving” in Europe | “10 Recommendations for Eco-driving” in Japan |
| 1. Anticipate traffic flow | - Release the accelerator earlier when decelerating |
| 2. Maintain a steady speed at low RPM | - Reduce acceleration and deceleration while keeping enough distance between cars |
| 3. Shift up early | - |
| 4. Check tire pressures frequently at least once a month and before driving at high speed | - Check the pressure of the tires as the first step toward better maintenance |
| 5. Consider any extra energy required costs fuel and money |
- Use air conditioners appropriately - Avoid unnecessary idling - Avoid traffic jams; leave home with time to spare - Take out unnecessary loads - Be aware of your fuel consumption |
| - |
- Press the accelerator gently when accelerating - Do not block traffic when parking |
|
1. Observe the speed limit and maintain a constant speed 2. Release the accelerator earlier when decelerating 3. Press the accelerator gently when accelerating |
| parameter | value |
| μ - | 0.013 |
| M [kg] | 1,110 |
| Cd - | 0.35 |
| S [m2] | 2.16 |
| δM [kg] | 38 |
| ECO-S | ECO-A | ECO-B | ECO-C | ||
| Travel Time (sec) | 648 | 628 | 627 | 612 | |
| Max Speed (km/h) | 56 | 64 | 66 | 80 | |
| Stopping Frequency | 5 | 5 | 6 | 5 | |
| Running Energy (kJ/km) | 277 | 316 | 395 | 463 | |
| Fuel Consumption (cc/km) | 45.3 | 48.5 | 59.7 | 67.5 | |
| Time Share | Idle | 14% | 20% | 26% | 28% |
| Cruise | 40% | 35% | 29% | 20% | |
| Acceleration | 23% | 24% | 26% | 24% | |
| Deceleration | 23% | 20% | 18% | 29% | |
| Vehicle Type | Engine Displacement | Transmission | Weight (nt.) | |
| 660cc CVT | ICEV | 1,800 cc | CVT | 770 kg |
| 1,300cc CVT | 1,300 cc | CVT | 1,000 kg | |
| 1,800cc 4AT | 660 cc | 4 AT | 1,170 kg | |
| 1,500cc HEV | HEV | 1,500 cc | e-CVT | 1,260 kg |
| EV Type A | EV | - | - | 1,030 kg |
| EV Type B | - | - | 1,070 kg |
