Lightweight Ship Object Detection Algorithm for Remote Sensing Images Based on Multi-scale Perception and Feature Enhancement
Wei Sun, Xinyi Shen, Xiaorui Zhang, Fei Guan · Photogrammetric Engineering & Remote Sensing · 2025
As global trade and maritime traffic develop, exploring ship detection in remote sensing images has become a research hotspot. However, ships in remote sensing images are so small that it leads to a high detection leakage rate and excessive model parameters, making them difficult to apply on remote sensing equipment with limited resources. To address the challenge, we propose a light-weight ship object detection algorithm, adaptive layered multi-scale You Only Look Once version 8 (ALM-YOLOv8), based on multi-scale perception and feature enhancement. To enhance the model's perception of contextual information in complex backgrounds, a multi-scale channel fusion module is constructed to extract features of various scales. To enhance the extracted features, a small-object detection layer and a dynamic channel attention convolution that assigns dynamic weights are proposed. Additionally, this study embeds the large separable kernel attention mechanism into the original network, which lightens the model. Experiments on the HRSC2016 dataset demonstrate the effectiveness of ALM-YOLOv8.