Single Shot Multibox Detector Network Research Articles

Model optimization and acceleration method based on meta-learning and model pruning for laser vision weld tracking system

Purpose This paper aims to propose a lightweight, high-accuracy object detection model designed to enhance seam tracking quality under strong arcs and splashes condition. Simultaneously, the model aims to reduce computational costs. Design/methodology/approach The lightweight model is constructed based on Single Shot Multibox Detector (SSD). First, a neural architecture search method based on meta-learning and genetic algorithm is introduced to optimize pruning strategy, reducing human intervention and improving efficiency. Additionally, the Alternating Direction Method of Multipliers (ADMM) is used to perform structural pruning on SSD, effectively compressing the model with minimal loss of accuracy. Findings Compared to state-of-the-art models, this method better balances feature extraction accuracy and inference speed. Furthermore, seam tracking experiments on this welding robot experimental platform demonstrate that the proposed method exhibits excellent accuracy and robustness in practical applications. Originality/value This paper presents an innovative approach that combines ADMM structural pruning and meta-learning-based neural architecture search to significantly enhance the efficiency and performance of the SSD network. This method reduces computational cost while ensuring high detection accuracy, providing a reliable solution for welding robot laser vision systems in practical applications.

Establishment and validation of an artificial intelligence-based model for real-time detection and classification of colorectal adenoma

Colorectal cancer (CRC) prevention requires early detection and removal of adenomas. We aimed to develop a computational model for real-time detection and classification of colorectal adenoma. Computationally constrained background based on real-time detection, we propose an improved adaptive lightweight ensemble model for real-time detection and classification of adenomas and other polyps. Firstly, we devised an adaptive lightweight network modification and effective training strategy to diminish the computational requirements for real-time detection. Secondly, by integrating the adaptive lightweight YOLOv4 with the single shot multibox detector network, we established the adaptive small object detection ensemble (ASODE) model, which enhances the precision of detecting target polyps without significantly increasing the model's memory footprint. We conducted simulated training using clinical colonoscopy images and videos to validate the method's performance, extracting features from 1148 polyps and employing a confidence threshold of 0.5 to filter out low-confidence sample predictions. Finally, compared to state-of-the-art models, our ASODE model demonstrated superior performance. In the test set, the sensitivity of images and videos reached 87.96% and 92.31%, respectively. Additionally, the ASODE model achieved an accuracy of 92.70% for adenoma detection with a false positive rate of 8.18%. Training results indicate the effectiveness of our method in classifying small polyps. Our model exhibits remarkable performance in real-time detection of colorectal adenomas, serving as a reliable tool for assisting endoscopists.

Floating object detection using double-labelled domain generalization

Detection of floating objects in complicated aquatic environments has a wide range of applications, but it confronts significant hurdles due to the imbalance in detection accuracy and efficiency, and low domain generalization performance. To address these issues, this study proposes a novel floating object detection method based on double-labelled domain generalization. First, the Single Shot Multibox Detector (SSD) is improved by replacing the backbone network with a lightweight feature extraction network, and dynamic feature pyramid network is introduced to balance accuracy and efficiency. Then, this study initializes the improved SSD network based on the double-labelled data of the source domain, and filters the feature extractor bias, classification bias and location bias using pseudo-labelling and feature projection based on the un-labelled source domain data to minimize the bias to improve the domain generalization performance. The proposed method is trained on a self-constructed floating object dataset and is compared with state-of-the-art methods based on multiple scenarios. The results show that the proposed method achieves better performance in double-labelled domain generalization and conventional domain generalization tasks compared to other methods, achieving 70.33%, 22.38 f/s and 85.29% and 17.81 f/s in accuracy and speed respectively, satisfying the need for multi-scale floating object detection in complex environments and also alleviating the data labelling problem. This work effectively solves the problem of slow detection of floating objects due to the complex model structure and low generalization ability, and provides support for the rapid detection of floating objects in complex scenarios and the promotion of technology applications.

A high-precision jujube disease spot detection based on SSD during the sorting process.

The development of automated grading equipment requires achieving high throughput and precise detection of disease spots on jujubes. However, the current algorithms are inadequate in accomplishing these objectives due to their high density, varying sizes and shapes, and limited location information regarding disease spots on jujubes. This paper proposes a method called JujubeSSD, to boost the precision of identifying disease spots in jujubes based on a single shot multi-box detector (SSD) network. In this study, a diverse dataset comprising disease spots of varied sizes and shapes, varying densities, and multiple location details on jujubes was created through artificial collection and data augmentation. The parameter information obtained from transfer learning into the backbone feature extraction network of the SSD model, which reduced the time of spot detection to 0.14 s. To enhance the learning of target detail features and improve the recognition of weak information, the traditional convolution layer was replaced with deformable convolutional networks (DCNs). Furthermore, to address the challenge of varying sizes and shapes of disease spot regions on jujubes, the path aggregation feature pyramid network (PAFPN) and balanced feature pyramid (BFP) were integrated into the SSD network. Experimental results demonstrate that the mean average precision at the IoU (intersection over union) threshold of 0.5 (mAP@0.5) of JujubeSSD reached 97.1%, representing an improvement of approximately 6.35% compared to the original algorithm. When compared to existing algorithms, such as YOLOv5 and Faster R-CNN, the improvements in mAP@0.5 were 16.84% and 8.61%, respectively. Therefore, the proposed method for detecting jujube disease spot achieves superior performance in jujube surface disease detection and meets the requirements for practical application in agricultural production.

Automatic Detection and Classification of Modic Changes in MRI Images Using Deep Learning: Intelligent Assisted Diagnosis System.

Modic changes (MCs) are the most prevalent classification system for describing intravertebral MRI signal intensity changes. However, interpreting these intricate MRI images is a complex and time-consuming process. This study investigates the performance of single shot multibox detector (SSD) and ResNet18 network-based automatic detection and classification of MCs. Additionally, it compares the inter-observer agreement and observer-classifier agreement in MCs diagnosis to validate the feasibility of deep learning network-assisted detection of classified MCs. A retrospective analysis of 140 patients with MCs who underwent MRI diagnosis and met the inclusion and exclusion criteria in Tianjin Hospital from June 2020 to June 2021 was used as the internal dataset. This group consisted of 55 males and 85 females, aged 25 to 89 years, with a mean age of (59.0 ± 13.7) years. An external test dataset of 28 patients, who met the same criteria and were assessed using different MRI equipment at Tianjin Hospital, was also gathered, including 11 males and 17 females, aged 31 to 84 years, with a mean age of 62.7 ± 10.9 years. After Physician 1 (with 15 years of experience) annotated all MRI images, the internal dataset was imported into the deep learning model for training. The model comprises an SSD network for lesion localization and a ResNet18 network for lesion classification. Performance metrics, including accuracy, recall, precision, F1 score, confusion matrix, and inter-observer agreement parameter Kappa value, were used to evaluate the model's performance on the internal and external datasets. Physician 2 (with 1 year of experience) re-labeled the internal and external test datasets to compare the inter-observer agreement and observer-classifier agreement. In the internal dataset, when models were utilized for the detection and classification of MCs, the accuracy, recall, precision and F1 score reached 86.25%, 87.77%, 84.92% and 85.60%, respectively. The Kappa value of the inter-observer agreement was 0.768 (95% CI: 0.656, 0.847),while observer-classifier agreement was 0.717 (95% CI: 0.589, 0.809).In the external test dataset, the model's the accuracy, recall, precision and F1 scores for diagnosing MCs reached 75%, 77.08%, 77.80% and 74.97%, respectively. The inter-observer agreement was 0.681 (95% CI: 0.512, 0.677), and observer-classifier agreement was 0.519 (95% CI: 0.290, 0.690). The model demonstrated strong performance in detecting and classifying MCs, achieving high agreement with physicians in MCs diagnosis. These results suggest that deep learning models have the potential to facilitate the application of intelligent assisted diagnosis techniques in the field of spine research.

Online Fabric Defects Detection Using Convolutional Neural Networks with Two Frameworks

Due to the suboptimal efficiency, accuracy, and increasing costs of manual defect detection in the textile industry, online visual inspection for fabric defects has emerged as an essential and promising research area. However, challenges such as the lack of defective samples and issues with industrial deployment still persist. This paper presents a novel defect detection technique based on deep learning, which primarily comprises two frameworks. First, we design an improved generative adversarial network with an encoder–decoder architecture to address the paucity of requisite defective samples. We use defect-free samples as input to the generator, ensuring that the generated defect samples maintain a similar pattern. We mitigate the vanishing gradient problem using Wasserstein distance as the loss function. Second, we enhance the Single Shot MultiBox Detector network by introducing Inception modules and feature fusion to detect defects across different scales. The AdaBound optimizer is selected to update the model parameters. We compare the proposed approach with other methods on self-generated fabric data sets that are partially produced by our generative adversarial network model. An online defect detection system is proposed to capture fabric images and evaluation in a production environment. Experiments demonstrate the superior performance of the proposed approach, achieving 97.5% accuracy in real time, making it well-suited for application in the industry.

Transfer Learning-Based Lightweight SSD Model for Detection of Pests in Citrus

In citrus cultivation, it is a difficult task for farmers to classify different pests correctly and make proper decisions to prevent citrus damage. This work proposes an efficient modified lightweight transfer learning model which combines the effectiveness and accuracy of citrus pest characterization with mobile terminal counting. Firstly, we utilized typical transfer learning feature extraction networks such as ResNet50, InceptionV3, VGG16, and MobileNetV3, and pre-trained the single-shot multibox detector (SSD) network to compare and analyze the classification accuracy and efficiency of each model. Then, to further reduce the amount of calculations needed, we miniaturized the prediction convolution kernel at the end of the model and added a residual block of a 1 × 1 convolution kernel to predict category scores and frame offsets. Finally, we transplanted the preferred lightweight SSD model into the mobile terminals developed by us to verify its usability. Compared to other transfer learning models, the modified MobileNetV3+RPBM can enable the SSD network to achieve accurate detection of Panonychus Citri Mcgregor and Aphids, with a mean average precision (mAP) up to 86.10% and the counting accuracy reaching 91.0% and 89.0%, respectively. In terms of speed, the mean latency of MobileNetV3+RPBM is as low as 185 ms. It was concluded that this novel and efficient modified MobileNetV3+RPBM+SSD model is effective at classifying citrus pests, and can be integrated into devices that are embedded for mobile rapid detection as well as for counting pests in citrus orchards. The work presented herein can help encourage farm managers to judge the degree of pest damage and make correct decisions regarding pesticide application in orchard management.

Deep Learning-Based Automatic Defect Detection Method for Sewer Pipelines

To address the issues of low automation, reliance on manual screening by professionals, and long detection cycles in current urban drainage pipeline defect detection, this study proposes an improved object detection algorithm called EFE-SSD (enhanced feature extraction SSD), based on the SSD (Single Shot MultiBox Detector) network. Firstly, the RFB_s module is added to the SSD backbone network to enhance its feature extraction capabilities. Additionally, multiple scale features are fused to improve the detection performance of small target defects to some extent. Then, the improved ECA attention mechanism is used to adjust the channel weights of the output layer, suppressing irrelevant features. Finally, the Focal Loss is employed to replace the cross-entropy loss in the SSD network, effectively addressing the issue of imbalanced positive and negative samples during training. This increases the weight of difficult-to-classify samples during network training, further improving the detection accuracy of the network. Experimental results show that EFE-SSD achieves a detection mAP of 92.2% for four types of pipeline defects: Settled deposits, Displaced joints, Deformations, and Roots. Compared to the SSD algorithm, the model’s mAP was increased by 2.26 percentage points—ensuring the accuracy of pipeline defect detection.

An Improved SSD Object Detection Algorithm Based on Attention Mechanism and Feature Fusion

The Single Shot MultiBox Detector (SSD) is a well-known object detection method, but its detection of small objects is not effective. This paper makes modifications to the SSD object detection method to address its insufficient semantic information in low-level feature maps, thus enhancing the detectability for small objects. First, the Feature Pyramid Network (FPN) is incorporated into the SSD so that the shallow feature map, which is primarily utilized for detecting small objects, contains more semantic information in addition to rich location information. Second, the Convolutional Block Attention Module (CBAM) is introduced to reinforce the SSD network’s capability to learn key features and thus improve missed detections. The experimental data indicate that this algorithm achieves 78.1% mAP in the PASCAL VOC2007test, a 3.9% improvement compared with the conventional SSD, and also has a great improvement compared with the Fast R-CNN and Faster R-CNN. As well as,this algorithm is better for small object detection and also meets the real-time requirements.

A Multiscale Polyp Detection Approach for GI Tract Images Based on Improved DenseNet and Single-Shot Multibox Detector

Small bowel polyps exhibit variations related to color, shape, morphology, texture, and size, as well as to the presence of artifacts, irregular polyp borders, and the low illumination condition inside the gastrointestinal GI tract. Recently, researchers developed many highly accurate polyp detection models based on one-stage or two-stage object detector algorithms for wireless capsule endoscopy (WCE) and colonoscopy images. However, their implementation requires a high computational power and memory resources, thus sacrificing speed for an improvement in precision. Although the single-shot multibox detector (SSD) proves its effectiveness in many medical imaging applications, its weak detection ability for small polyp regions persists due to the lack of information complementary between features of low- and high-level layers. The aim is to consecutively reuse feature maps between layers of the original SSD network. In this paper, we propose an innovative SSD model based on a redesigned version of a dense convolutional network (DenseNet) which emphasizes multiscale pyramidal feature maps interdependence called DC-SSDNet (densely connected single-shot multibox detector). The original backbone network VGG-16 of the SSD is replaced with a modified version of DenseNet. The DenseNet-46 front stem is improved to extract highly typical characteristics and contextual information, which improves the model's feature extraction ability. The DC-SSDNet architecture compresses unnecessary convolution layers of each dense block to reduce the CNN model complexity. Experimental results showed a remarkable improvement in the proposed DC-SSDNet to detect small polyp regions achieving an mAP of 93.96%, F1-score of 90.7%, and requiring less computational time.

Deep Learning Based Vehicle Speed Estimation on Highways

Traffic management always is a matter requiring the attention of highway system managers in terms of vehicle monitoring and speed estimation. This paper proposes an efficient deep learning-based vehicle speed estimation on highway lanes in the Vietnam transport system. The input videos are recorded by fixed surveillance cameras. An optimized single shot multibox detector network, called SSD, is utilized for vehicle license plate detection (LPD). The deep SORT (simple online and real-time tracking) model is first applied to video vehicle tracking and performed in the detected license plate area. This tracking process investigates the traveling distance of a vehicle to estimate its speed. In this study, the dataset has been normalized to improve the efficiency of vehicle localization and tracking to improve the time elapsing in the estimation of the distance traveled by vehicles on highways. The results showed that the proposed system has achieved better accuracy in terms of the determined speeds with errors ranging between [-1.5, +1.1] km/h, equivalent to 98% of the error limit by the regulation in Viet Nam.

ISHD: Intelligent Standing Human Detection of Video Surveillance for the Smart Examination Environment

In the environment of smart examination rooms, it is important to quickly and accurately detect abnormal behavior (human standing) for the construction of a smart campus. Based on deep learning, we propose an intelligent standing human detection (ISHD) method based on an improved single shot multibox detector to detect the target of standing human posture in the scene frame of exam room video surveillance at a specific examination stage. ISHD combines the MobileNet network in a single shot multibox detector network, improves the posture feature extractor of a standing person, merges prior knowledge, and introduces transfer learning in the training strategy, which greatly reduces the computation amount, improves the detection accuracy, and reduces the training difficulty. The experiment proves that the model proposed in this paper has a better detection ability for the small and medium-sized standing human body posture in video test scenes on the EMV-2 dataset.

Improved SSD network for fast concealed object detection and recognition in passive terahertz security images

With the strengthening of global anti-terrorist measures, it is increasingly important to conduct security checks in public places to detect concealed objects carried by the human body. Research in recent years has shown that deep learning is helpful for detecting concealed objects in passive terahertz images. However, previous studies have failed to achieve superior accuracy and performance for real-time labeling. Our research aims to propose a novel method for accurate and real-time detection of concealed objects in terahertz images. To reach this goal we trained and tested a promising detector based on deep residual networks using human image data collected by passive terahertz devices. Specifically, we replaced the backbone network of the SSD (Single Shot MultiBox Detector) algorithm with a more representative residual network to reduce the difficulty of network training. Aiming at the problems of repeated detection and missed detection of small targets, a feature fusion-based terahertz image target detection algorithm was proposed. Furthermore, we introduced a hybrid attention mechanism in SSD to improve the algorithm’s ability to acquire object details and location information. Finally, the Focal Loss function was introduced to improve the robustness of the model. Experimental results show that the accuracy of the SSD algorithm is improved from 95.04 to 99.92%. Compared with other current mainstream models, such as Faster RCNN, YOLO, and RetinaNet, the proposed method can maintain high detection accuracy at a faster speed. This proposed method based on SSD achieves a mean average precision of 99.92%, an F1 score of 0.98, and a prediction speed of 17 FPS on the validation subset. This proposed method based on SSD-ResNet-50 can provide a technical reference for the application and development of deep learning technology in terahertz smart security systems. In the future, it can be widely used in some public scenarios with real-time security inspection requirements.

Individual identification of cashmere goats via method of fusion of multiple optimization

AbstractFacial recognition technology and related research have matured over time, but research in the field of individual animal recognition is still very limited. Therefore, this article focuses on the identification of cashmere goats with similar characteristics. First, the single shot multibox detector network was used to process the dataset. Next, transfer learning was applied to learn the characteristics of the goats, as well as the loss function is composed of Triplet Loss and Label Smoothing CrossEntropy Loss function. The result of Label Smoothing CrossEntropy Loss function is fused by multiple different branches, which is convenient for classification. We added a small number of images of 24 different breeds of sheep to each cashmere goat dataset with different ID to promote the distance between training individuals, and then used the trained model to find the number of goats with the lowest recognition accuracy. The Cycle‐Consistent Adversarial Network (Cycle‐GAN) learned the goat dataset with a high error rate in individual identification. Unlike previous studies using the Cycle‐GAN, we took the novel approach of using this network to learn and combine the features seen in photos of cashmere goats. Since the learned features were all observed in the same goats, this method achieved better results in learning the features of the goats. Finally, we found that recognition can be performed on this data with an accuracy of 93.75%. These results suggest that identification based on deep learning has a high accuracy rate, as well as great value in identifying individual cashmere goats.

(Retracted) Target recognition and localization based on lightweight single-shot multibox detector network for robotics

Target recognition and localization are essential in computer vision and pattern recognition in robotics. The artificial extraction of features was omitted with the emergence of deep convolutional neural networks, reducing the influence of human factors on the results. The single-shot multibox detector (SSD) network has achieved excellent recognition by high precision and fast speed in target recognition and positioning. However, some small real-time systems are difficult to implement because of their demanding hardware and extended training time. Based on the digital normalization and residual network structure, the depth-wise separable convolution is proposed to replace the traditional convolution. The improved SSD network structure was used for identification and positioning in our work. The speed of training and testing increased without a decline in the accuracy, thus reducing the dependence on hardware. The method has achieved good results on the PASCAL VOC dataset after testing. It can also be applied to the field of intelligent inspection robots and intelligent security robots.

Single Shot Multibox Detector Automatic Polyp Detection Network Based on Gastrointestinal Endoscopic Images.

Purpose In order to resolve the situation of high missed diagnosis rate and high misdiagnosis rate of the pathological analysis of the gastrointestinal endoscopic images by experts, we propose an automatic polyp detection algorithm based on Single Shot Multibox Detector (SSD). Method In the paper, SSD is based on VGG-16, the fully connected layer is changed to a convolutional layer, and four convolutional layers with successively decreasing scales are added as a new network structure. In order to verify the practicability, it is not only compared with manual polyp detection but also with Mask R-CNN. Results Multiple experimental results show that the mean Average Precision (mAP) of the SSD network is 95.74%, which is 12.4% higher than the manual detection and 5.7% higher than the Mask R-CNN. When detecting a single frame of image, the detection speed of SSD is 8.41 times that of manual detection. Conclusion Based on the traditional pattern recognition algorithm and the target detection algorithm using deep learning, we select a variety of algorithms to identify and classify polyps to achieve efficient detection results. Our research demonstrates that deep learning has a lot of room for development in the field of gastrointestinal image recognition.

Recognition of the internal situation of aircraft skin based on deep learning

The aircraft skin is an important component of the aircraft, and its integrity affects the flight performance and safety performance of the aircraft. Damage detection technology with ultrasonic nondestructive testing as the core has played an important role in aircraft skin damage detection. Due to the difficulty in aircraft skin detection, relying solely on the ultrasonic A-scan equipment has very low detection efficiency. The introduction of artificial intelligence can effectively improve the detection efficiency. This paper establishes the one-dimensional convolutional neural network and single shot multibox detector network, which is based on the SSD network and uses dilated convolution to improve the real-time tracking of the ultrasonic probe. At the same time, 1DCNN is introduced to classify the ultrasonic A-scan signal. Finally, the detection results of the two are combined to achieve the reflection of the internal conditions of the aircraft skin. After testing, the algorithm can identify skin simulation specimens, and its recognition accuracy is 96.5%, AP is 90.9%, and kappa is 0.996. Comparing the improved SSD network with networks such as SSD, YOLOv3, and Faster R-CNN, the results show that the improved network used in this paper is more excellent and effective. At the same time, the detection effects of four types of optimization algorithms and five learning rates are studied, and finally, the corresponding signal classification model adopts Adam and the learning rate of 0.0001 has the best effect.

A fast workpiece detection method based on multi-feature fused SSD

PurposeWorkpiece sorting is a key link in industrial production lines. The vision-based workpiece sorting system is non-contact and widely applicable. The detection and recognition of workpieces are the key technologies of the workpiece sorting system. To introduce deep learning algorithms into workpiece detection and improve detection accuracy, this paper aims to propose a workpiece detection algorithm based on the single-shot multi-box detector (SSD).Design/methodology/approachPropose a multi-feature fused SSD network for fast workpiece detection. First, the multi-view CAD rendering images of the workpiece are used as deep learning data sets. Second, the visual geometry group network was trained for workpiece recognition to identify the category of the workpiece. Third, this study designs a multi-level feature fusion method to improve the detection accuracy of SSD (especially for small objects); specifically, a feature fusion module is added, which uses “element-wise sum” and “concatenation operation” to combine the information of shallow features and deep features.FindingsExperimental results show that the actual workpiece detection accuracy of the method can reach 96% and the speed can reach 41 frames per second. Compared with the original SSD, the method improves the accuracy by 7% and improves the detection performance of small objects.Originality/valueThis paper innovatively introduces the SSD detection algorithm into workpiece detection in industrial scenarios and improves it. A feature fusion module has been added to combine the information of shallow features and deep features. The multi-feature fused SSD network proves the feasibility and practicality of introducing deep learning algorithms into workpiece sorting.

Enhanced SSD with interactive multi-scale attention features for object detection

Single Shot MultiBox Detector (SSD) method using multi-scale feature maps for object detection, showing outstanding performance in object detection task. However, as a one-stage detection method, it’s difficult for SSD methods to quickly notice significant areas of objects in the image. In the SSD network structure, feature maps of different scales are used to independently predict object, and there is a lack of interaction between low-level feature maps and high-level feature maps. In this paper we propose an enhanced SSD method using interactive multi-scale attention features (MA-SSD). Our method uses the attention mechanism to generate attention features of multiple scales and adds it to the original detection branch of the SSD method, which effectively enhances the feature representation ability and improves the detection accuracy. At the same time, the feature of different detection scales interacts with each other, and all the detection branches in our method have a parallel structure, which ensures the detection efficiency. Our proposed method achieves competitive performance on the public dataset PascalVOC.

Automated Asphalt Highway Pavement Crack Detection Based on Deformable Single Shot Multi-Box Detector Under a Complex Environment

Pavement cracks are severely affecting highway performance. Thus, implementing high-precision highway pavement crack detection is important for highway maintenance. However, the asphalt highway pavement environment is complex, and different pavement backgrounds are more difficult than others for detecting highway pavement cracks. Interference from road markings and surface repairs also complicate the environments and thus the detection of crack. To reduce interference, we collected many images from different highway pavement backgrounds. We also improved the single shot multi-box detector (SSD) network and proposed a novel network named deformable SSD by adding a deformable convolution to the backbone feature extraction network VGG16. We verified our model using the PASCAL VOC2007 dataset and obtained a mean average precision ( mAP ) 3.1% higher than that of the original SSD model. We then trained and tested the proposed model using our crack detection dataset. We calculated precision, recall, F1 score, AP, mAP , and FPS to examine the performance of our model. The mAP of all categories in the test data was 85.11% using the proposed model 10.4% and 0.55% more than that of YOLOv4 and the original SSD model, respectively. These findings show that our model outperforms YOLOv4 and the original SSD model and confirm that incorporating a deformable convolution into the SSD network can improve the model’s performance. The proposed model is appropriate for detecting pavement crack categories and locations in complicated environments. It can also provide important technical support for highway maintenance.