MAL-YOLO: a lightweight algorithm for target detection in side-scan sonar images based on multi-scale feature fusion and attention mechanism

Side-scan sonar image target detection is of great significance in seabed resource exploration and other fields. However, affected by the complex underwater environment, side-scan sonar images have the problems of few target samples and large differences in the scale of each type of target. In addition, the computational complexity of high-performance models based on deep learning is too high to be applied on platforms with limited computational resources. To solve these problems, this paper proposes a lightweight algorithm for target detection in side-scan sonar images based on multi-scale feature fusion and attention mechanism (MAL-YOLO). Firstly, a lightweight feature extraction module is used. This module combines depthwise separable convolution and efficient multi-scale attention (EMA) module to improve the feature extraction capability of the model while reducing the computational volume. Secondly, a multi-scale feature fusion network combining asymptotic feature pyramid network and EMA module is used to enhance the fusion and representation of multi-scale features in the model. Finally, the MPDIoU loss function is used to provide more accurate bounding box regression. The experimental results show that the algorithm has significant advantages in both detection accuracy and model lightweighting compared with the current state-of-the-art algorithms such as YOLOv7 and YOLOv8.