CCP-YOLO: Improved visible ship target recognition for YOLOv7-tiny

Aiming at the problems of similar target features in visible light ship images, susceptible to factors such as light and weather changes, and low model recognition accuracy due to large changes in target scale, this paper proposes an improved YOLOv7-tiny visible light ship target recognition model, CCP-YOLO. Based on the CBAM (Convolutional Block Attention Module) attention mechanism, the channel attention and spatial attention are concatenated, and a multi-scale convolutional layer is added at the front end of the channel attention to form the feature enhancement module CBAM+, which enhances the model's ability to focus on important areas and improves the differentiation of different ship types. Additionally, the feature extraction module CCC (Convolution-Convolution-CBAM+) is designed by combining CBAM+ with ordinary convolution operations using multi-branch concatenation and residual connections, which further strengthens feature extraction capabilities, effectively suppresses noise in complex backgrounds, and reduces the interference of light and weather changes. Finally, the PANet (Path Aggregation Network) in YOLOv7-tiny is enhanced to more accurately focus on targets of different scales using cross-layer connectivity, thereby refining the recognition process. With these improvements, the model significantly enhances detection accuracy compared to the baseline YOLOv7-tiny, achieving an increase of 2.5% in [email protected] and 1.7% in [email protected]:0.95 in tests on the Singapore Maritime dataset.