AFQSeg: An Adaptive Feature Quantization Network for Instance-Level Surface Crack Segmentation

It is critical to deploy complicated neural network models on hardware with limited resources. This paper proposes a novel model quantization method, named the Low-Cost Proxy-Based Adaptive Mixed-Precision Model Quantization (LCPAQ), which contains three key modules. The hardware-aware module is designed by considering the hardware limitations, while an adaptive mixed-precision quantization module is developed to evaluate the quantization sensitivity by using the Hessian matrix and Pareto frontier techniques. Integer linear programming is used to fine-tune the quantization across different layers. Then the low-cost proxy neural architecture search module efficiently explores the ideal quantization hyperparameters. Experiments on the ImageNet demonstrate that the proposed LCPAQ achieves comparable or superior quantization accuracy to existing mixed-precision models. Notably, LCPAQ achieves 1/200 of the search time compared with existing methods, which provides a shortcut in practical quantization use for resource-limited devices.

An automated road defect detection system is a key part of intelligent traffic infrastructure maintenance. Existing object detection models have slow inference speed and low detection accuracy. This problem is more serious under shadows, oil stains, and occlusion from natural environments. Detection accuracy of existing models drops sharply in complex environments. This paper proposes a lightweight and high-accuracy crack detection framework based on YOLOv8s, named Crack-YOLO. It is designed to solve slow speed and low accuracy in road defect detection under complex conditions. The model replaces the original convolution modules in YOLOv8 with Context-Guided (CG) modules. It uses the designed C2f_DynamicConv to replace static convolution kernels. The original detection head is replaced with an Adaptive Spatial Feature Fusion (ASFF) head. Experiments and analysis are conducted on four datasets: CrackVariety, CrackTree200, Crack500, and CFD. Experimental results show that Crack-YOLO greatly outperforms YOLOv8s in both detection speed and detection accuracy. On the self-built CrackVariety dataset, the mAP@0.5 reaches 71.4%. The inference speed reaches 416 FPS. Compared with the baseline YOLOv8s model, accuracy increases by 31.0%. The speed increases by nearly two times. In addition, the model is successfully deployed on a Raspberry Pi 5 edge device. It is integrated with the ASTM D6433 standard. After detecting road defects, the Pavement Condition Index (PCI) is calculated automatically. This proves its practicality on devices with limited computing resources.

SA-MVSNet: Self-attention-based multi-view stereo network for 3D reconstruction of images with weak texture

Crack detection has always been a dominant requirement for steel industries to ensure quality production and seamless infrastructure maintenance. However, application complexities and defect morphological differences make existing approaches confined. Steel-strip surface often undergoes scratch, crack and fatigue conditions during production. Manual crack detection schemes are no longer effective in current day complex environment. Amongst major steel strip crack detection approaches vision based techniques have been found potential. Filamentous crack which is caused due to fatigue or strain is fine-grained and thin and hence highly difficult to be detected by classical morphology and static threshold based schemes. In the present work steel strip surface crack(filamentous) detection system is developed which employs Varying-Morphological Segmentation (VMS) also called Neuron-Model Segmentation (NMS) in conjunction with local directive filtering and active contour propagation. The proposed method can be stated as an augmented variational framework that employs multi-directional filters for local crack-region identification followed by automated multi-directional region growing and iterative contour evolution which performs level set energy minimization to achieve accurate crack detection even under topological non-linearity and varying illumination conditions .Simulation results with standard benchmark data has confirmed that the proposed method exhibits satisfactory performance for steel strip surface cracks.

Bridge crack detection is a key task in the structural health monitoring of Civil Engineering. In the traditional bridge crack detection methods, there exist some problems such as high cost, low speed, and complex structure. This paper developed a bridge surface crack detection system based on improved YOLOv5s. The GhostBottleneck module was employed to replace the classic C3 module of the YOLOv5s backbone network, meanwhile the channel attention module namely ECA-Net was also added to the network, which not only reduced the amount of calculation, but also enhanced the ability of the network in extracting cross-channel information features. The adaptive spatial feature fusion (ASFF) was introduced to address the conflict problem caused by the inconsistency of feature scale in the network feature fusion stage, and the transfer learning was utilized to train the network. The experimental results showed that the improved YOLOv5s performed better than Faster R-CNN, SSD, YOLOv3, and YOLOv5s, with the Precision of 93.6%, Recall of 95.4%, and mAP of 98.4%. Further, the improved YOLOv5s was deployed in PyQt5 to realize the real-time detection of bridge cracks. This research showed that the proposed model not only provides a novel solution for bridge surface crack detection, but also has certain industrial application value.

The detection and prevention of concrete cracks and spalls is crucial to ensure the structural integrity and longevity of civil infrastructure. In this research paper, we propose a novel method for concrete crack and spall detection based on YOLOv8 with ByteTrack and supervision. The proposed method exploits the advantages of the YOLOv8 object detection framework, which provides real-time and accurate detection, tracking and counting of various objects. By integrating ByteTrack, a state-of-the-art tracking algorithm, we increase system performance and robustness in tracking cracks and spalls over time. Furthermore, supervision is employed to improve detection accuracy through iterative training and fine-tuning. To train the model, an extensive dataset of concrete crack and spall images is collected, annotated, and enhanced. The dataset includes a variety of scenarios, lighting conditions, and crack/spill sizes, which ensures the model's ability to generalize to real-world situations. Transfer learning is used to support the YOLOv8 backbone with pre-trained weights, to accelerate the convergence of the training process. Experimental evaluation is conducted on a benchmark dataset, and the proposed method outperforms existing techniques in terms of accuracy, precision, and recall. YOLOv8 with ByteTrack and Supervision achieves an overall accuracy of 94% detection rate for concrete cracks and spalls, even under challenging conditions. Real-time deployment is achieved on a high-performance computing platform, allowing efficient and timely monitoring of concrete structures. The proposed method demonstrates its potential as a valuable tool for infrastructure management and maintenance, enabling early detection of concrete cracks and spalls. By promptly identifying such defects, necessary repair and reinforcement measures can be implemented, preventing further deterioration and ensuring the safety and longevity of civil infrastructure. The high detection performance of the proposed approach, achieving excellent precision, recall, and F1 score for both crack and spall detection. Our system's overall average precision of 93% indicates its accuracy in detecting concrete cracks and spalls. Additionally, the system exhibits fast real-time processing with an inference speed of 50fps and is highly robust in handling diverse concrete crack and spall scenarios. Future work may focus on expanding the dataset, exploring the integration of additional detection algorithms, and evaluating system performance in large-scale applications.

While adapting pretrained vision models to downstream dense prediction tasks is widely used, current methods often overlook adaptation efficiency, especially in the context of multi-task learning (MTL). Although parameter-efficient fine-tuning (PEFT) methods can enhance parameter efficiency, broader aspects such as GPU memory and training time efficiency remain underexplored. In this paper, we propose a new paradigm that simultaneously achieves efficiency in Parameters, GPU Memory, and Training Time for Multi-Task Dense Vision Adaptation. Specifically, we propose a dual-branch framework, in which a frozen pretrained backbone serves as the generic main branch, and the proposed Bi-Directional Task Adaptation (BDTA) modules are integrated in parallel to form a task bypass branch that extracts adaptation features required by multiple specific tasks. This adaptation module is lightweight, efficient, and does not require backpropagation through the large pre-trained backbone, thus avoiding resource-intensive gradient computations. Moreover, a Mixture of Task Experts mechanism (MoTE) is further proposed to integrate adaptation features across tasks and scales, thereby obtaining more robust representations tailored for dense prediction tasks. On the PASCAL-Context benchmark, our method achieves over 2× relative performance improvement compared to the best prior multi-task PEFT method, while using only ~30% of the parameters, ~50% of the memory, and ~60% of the training time, demonstrating superior overall adaptation efficiency.

YOLOv5s-road: Road surface defect detection under engineering environments based on CNN-transformer and adaptively spatial feature fusion

Construction of an industrial detection model for tomato organic acids based on low-rank adaptation and spatial feature fusion

Underwater cracks are difficult to detect and observe, posing a major challenge to crack detection. Currently, deep learning-based underwater crack detection methods rely heavily on a large number of crack images that are difficult to collect due to their complex and hazardous underwater environments. This study proposes a new underwater image-processing method that combines a novel white balance method and bilateral filtering denoising method to transform underwater crack images into high-quality above-water images with original crack features. Crack detection is then performed based on an improved YOLOv9-OREPA model. Through experiments, it is found that the new image-processing method proposed in this study significantly improves the evaluation indicators of new images, compared with other methods. The improved YOLOv9-OREPA also exhibits a significantly improved performance. The experimental results demonstrate that the method proposed in this study is a new approach suitable for detecting underwater cracks in dams and achieves the goal of transforming underwater images into above-water images.

Automatic real-time crack detection using lightweight deep learning models

Automated high-precision crack detection in asphalt pavements is vital for road safety and infrastructure longevity. However, existing RGB-based models are often affected by environmental interference, and limited training datasets hinder their generalisation. To address these issues, this study presents a novel multimodal crack segmentation method (FC_SAM), utilising fine-tuned large vision models. Firstly, the method integrates infrared and RGB images through automatic registration and fusion to enhance crack detection. Then, two parameter-efficient fine-tuning (PEFT) techniques optimise the model. Finally, a data quantity test identifies the minimum dataset size for optimal performance. Besides, cross-device generalisation is validated using data from two infrared devices. Experimental results show that with only 300 training samples, the model achieves excellent segmentation (IoU: 86.41%, F1-score: 92.55%) and generalisation (IoU: 64.29%, F1-score: 78.10%), outperforming existing models. This research offers a novel, cost-effective, and robust solution for automated asphalt crack detection.

Cervical cancer is one of the most common malignant tumors, with high incidence and mortality rates. Recent studies mainly adopt Artificial Intelligence (AI) models to detect cervical cells. Yet, due to the imperceptible symptoms of cervical cells, there are three problems that may hinder the performance of the existing approaches: (a) poor quality of the whole-slide image (WSI) performed on cervical cells may lead to undesirable performance; (b) several types of abnormal cervical cells are involved in the progression of cervical cells from normal to cancer, which requires extensive clinical data for training; and (c) the diagnosis of the WSI is medical-rule-driven and requires the AI model to provide interpretability. To address these issues, we propose an integrated automatic cervical cancer screening (IACCS) framework. First, the IACCS framework incorporates a quality assessment module utilizing binarization-based cell counting and a Support Vector Machine (SVM) approach to identify fuzzy regions, ensuring WSI suitability for analysis. Second, to overcome the data limitations, the framework employs data enhancement techniques alongside incremental learning (IL) and active learning (AL) mechanisms, allowing the model to adapt progressively and learn efficiently from new data and expert feedback. Third, recognizing the need for interpretability, the diagnostic decision process is modeled as a multi-objective optimization problem. A multi-objective optimization algorithm is used to generate a set of interpretable diagnostic rules that offer explicit trade-offs between sensitivity and specificity. Extensive experiments demonstrate the effectiveness of the proposed IACCS framework. Applying our comprehensive framework yielded significant improvements in detection accuracy, achieving, for example, a 6.34% increase in mAP50:95 compared to the baseline YOLOv8 model. Furthermore, the generated Pareto-optimal diagnostic rules provide superior and more flexible diagnostic options compared to traditional manually defined rules. This research presents a validated pathway towards more robust, adaptable, and interpretable AI-assisted cervical cancer screening.

Dams are critical hydraulic infrastructures whose structural integrity directly impacts service safety and lifespan. Accurate detection of structural cracks is therefore essential for maintaining their stable and reliable operation. However, existing crack detection methods often exhibit high omission and false segmentation rates for fine cracks, which compromise detection accuracy. In addition, the algorithms are typically complex, resulting in low computational efficiency and difficulty in meeting real-time detection requirements. To address these limitations, we propose an optimized U2Net-based model named LAFEN. From the geometric perspective, a Deformable Convolutional Network (DCN) is employed to adaptively capture irregular crack shapes. From the frequency perspective, a Frequency-Domain Attention Module (FAM) is designed to jointly model geometric structures and high-frequency texture information. From the multi-scale perspective, an Adaptive Feature Aggregation Module (AFAM) is introduced to enhance the sensitivity and robustness to multiscale geometric features. These three modules are integrated into an Adaptive Feature-Enhancement Module (AFEM) to strengthen the capability of the U2Net for fine-crack feature extraction. To achieve a lightweight architecture, the RSU-7 module of U2Net is removed. Comprehensive experimental evaluations on a self-constructed dataset demonstrates that LAFEN achieves superior performance across multiple metrics, with an F1 score of 0.7903 and an IoU of 0.6698, while reducing the number of parameters by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 2. 7 \%}$</tex> and inference time by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 9 \%}$</tex>. These results indicate that LAFEN effectively balances detection accuracy, computational efficiency, and lightweight design, making it wellsuited for real-time dam crack detection applications.

End-to-end deep learning model for underground utilities localization using GPR