Instance segmentation is a task that involves pixel‐level classification and segmentation of each object instance in images. Various CNN‐based methods have achieved promising results in natural image instance segmentation. However, the noise interference, low resolution, and blurred edges bring more significant challenges for sonar image instance segmentation. To solve these problems, we propose the Effective Strategy for Sonar Images Instance Segmentation (ESSIIS). We introduce ASception, a new network combining Atrous Spatial Pyramid Pooling (ASPP) and Extreme Inception (Xception). By integrating this with ResNet and transforming traditional convolutions into deformable convolutions, we further improve the ability of the network to extract features from sonar images. Additionally, we incorporate a bidirectional feature fusion module to enhance information fusion. Finally, we evaluate the detection accuracy and segmentation accuracy of the proposed method on the public sonar image dataset and the self‐constructed dataset. ESSIIS attains a detection accuracy of 0.981 and a segmentation accuracy of 0.951 on SCTD, further impressively achieving 0.986 in both metrics when appraised on our dataset. The evaluation results demonstrate that the proposed method is more accurate, robust, and considerable for sonar image detection and segmentation.

Sonar images possess characteristics such as low resolution, high noise, and blurred edges. Utilizing CNNs would lead to problems such as inadequate target recognition accuracy. Moreover, due to their larger sizes and higher computational requirements, existing CNNs face deployment issues in embedded devices. Therefore, we propose a sonar image recognition algorithm optimized for the lightweight algorithm, MobileViT, by analyzing the features of sonar images. Firstly, the MobileViT block is modified by adding and redesigning the jump connection layer to capture more important features of sonar images. Secondly, the original 1 × 1 convolution is replaced with the redesigned multi-scale convolution Res2Net in the MV2 module to enhance the ability of the algorithm to learn global and local features. Finally, the IB loss is applied to address the imbalance of sample categories in the sonar dataset, assigning different weights to the samples to improve the performance of the network. The experimental results show that several proposed improvements have improved the accuracy of sonar image recognition to varying degrees. At the same time, the proposed algorithm is lightweight and can be deploy on embedded devices.

The railway track, as a fundamental infrastructure for rail transport, directly impacts the safety of train operations. Track defects such as cracks and bolt loosening, if not detected and repaired in time, can lead to severe safety accidents. In recent years, deep learning-based track defect detection algorithms have achieved good results. However, there are still some issues that need further research, such as low detection accuracy in complex track scenarios and limited defect detection capability for diverse railway components. To address these problems, this paper proposes an improved algorithm based on RTMDet. Firstly, an enhanced PSA polarized self-attention mechanism and Normlincomb activation function are added to the feature extraction module of RTMDet to enhance the model's feature representation capability. Secondly, a custom CSR-PAFPN feature fusion module is introduced to more effectively combine information of different scales, improving the model's performance in detecting various defects. Additionally, a detection head for small-sized track defects is added, utilizing high-resolution feature maps for more accurate detection of minute defects. Finally, a custom dataset for track defect detection is constructed, and the proposed method is tested on this dataset. Experimental results show that on the custom dataset, the improved algorithm achieved mAP of 66%, which is an increase of 3.5% over the original RTMDet algorithm and also higher than other outstanding object detection algorithms.