Image dehazing is a research focus, however the existing methods do not make enough use of depth information, which leads to poor dehazing effect for large depth scenes. Therefore, according to the relevance between the difference of haze concentration and depth in different regions of the image, it has become a challenging task. To address this challenge, firstly, we propose a novel multiscale depth information fusion enhancement network to improve dehazing ability in scenes with large depth changes. Secondly, our method hierarchically fuses the multiscale features of the depth map with the hazy image. We built a grade-by-grade fusion structure to fuse and enhance contextual features with attention mechanism. Finally, to make our model better adapt to real hazy images, restore the image while maintaining details. We add local prior knowledge supervision to each branch. We compare our model with other state-of-the-art methods on hazy datasets. The experimental results show that our dehazing model not only achieve optimal performance on synthetic images but also show certain advantages in real-world hazy images.

Crack detection is a fundamental effort to ensure road driving safety, aiming to detect potential safety hazards and avoid serious accidents. However, cracks can not be extracted completely and accurately due to problems such as low contrast, high noise, and complex topology of pavement cracks. To address these issues, we propose a directional connectivity feature enhancement network for pavement crack detection. In this network, we build multi-directional enhanced convolution to capture the complex topology of cracks, which is more sensitive to long cracks. To leverage both low-level detail information and high-level semantic information in the network, a novel multi-scale fusion attention is constructed to strengthen the mutual guidance of crack information between channels. The directional connectivity is introducted to establish loss module, which enhances the position and direction information between neighbouring pixels and further refines the crack edge features. To validate the effectiveness and accuracy of the proposed method, we experiment on six publicly available crack datasets, DeepCrack, Crack500, CFD, DCD, EdmCrack600 and DCCE. Compared to other networks, our network achieves 2.2% improvement in ODS and 1.8% improvement in MIoU on the DeepCrack dataset, and 1.2% improvement in ODS and 0.9% improvement in MIoU on the Crack500 dataset. Sufficient experimental results show that our network has better crack detection performance.

Due to the existence of domain shift, the detection accuracy of road damage detection will drop significantly when the training images and test images are from different scenes. To address this problem, we propose a domain-invariant feature enhancement domain adaptation method (DIFEDA) based on the You Only Look Once (YOLO) series object detectors. We integrate three domain-invariant feature decoupling (DIFD) modules on the backbone to decouple multi-scale domain-invariant features through two-stage adversarial learning. The decoupled features are fed back to the backbone to realize domain-invariant feature enhancement. We construct a spatial and frequency domain perception (SPFD) module in the DIFD to decouple local and global domain-invariant features from the spatial level and frequency level, respectively. We also design a region segmentation decoder (RSD) to make the DIFD pay more attention to the domain-invariant feature extraction in the damaged region, thereby suppressing the interference of background information. We apply DIFEDA to YOLOv8, YOLOv9, YOLO11, and YOLOv12, and conduct extensive experiments in three cross-scenes. The experimental results show that DIFEDA can significantly improve the performance of all baselines, with up to 8.5% and 9.8% improvement in mAP@0.5 and F1, respectively, proving the effectiveness and generalization of our method.

Cracks are the most common disease in road maintenance work. They will accelerate the aging and damage of the pavement and bring potential risks to driving safety. Automatic detection of pavement cracks ensures infrastructure road safety. However, automatic detection of pavement cracks has always faced many challenges, such as irregular crack shapes and background noise interference. The existing methods ignore the correlation between cracked pixels, and there are some problems such as loss of detail information, lack of feature fusion and semantic gap. A Multi-Attention Fusion Dual-Branch Network(MFDB-Net) is proposed to accurately detect complete cracks. In this network, a dual-branch network is used as the backbone. A deeper branch extracts the semantic information of cracks, and a shallower branch retains more crack detail information. In addition, multiscale features are extracted by connecting multiple dilated convolutions in series with different convolution kernel sizes, which establishes global context connections. The attention mechanism is introduced to extract high-level network features and low-level network features respectively. Fusing features brings into play the characteristics of different network layers. Five publicly available crack datasets are used in our experiments, and the assessment is performed based on six evaluation metrics. We achieve F1-score of 0.879 and MIoU of 0.886 on the DeepCrack, F1-score of 0.737 and MIoU of 0.775 on the Crack500, F1-score of 0.571 and MIoU of 0.698 on the CrackTree200, and F1-score of 0.602 and MIoU of 0.712 on the EdmCrack600. We also carry out the generalization experiments on CFD. Sufficient experiments demonstrate our method present superior performance compared to other methods.