Improved YOLOv8 for effective ship detection in occluded backgrounds

Accurate ship object detection is crucial for navigational safety and effective maritime traffic management. Existing ship object detection models are often affected by variations in ship scale and occlusion, leading to frequent missed detections and false positives, making it challenging to achieve both high accuracy. To address these issues, this paper introduces an improved ship object detection model based on YOLOv8. First, the ReLU activation function in SEConv is replaced with the Swish activation function to enhance the model's nonlinear representation capability and its ability to handle complex backgrounds. Then, the improved SEConv is used to replace the standard convolution in C2f, forming the C2fSE module, which balances inference speed and accuracy, which is crucial for real-time object detection systems. Finally, CBAM is introduced before SPPF and C2f to concentrate the model's attention mechanism, emphasizing features that can predict ship occlusions, thus improving detection accuracy. Experimental results showed that the [email protected] increased by 3.66% and [email protected]:.95 increased by 7.89%, while reducing the number of parameters by 2.1M, effectively balancing model parameter and detection accuracy. Especially in complex scenarios involving ship occlusions, the improved model demonstrates higher accuracy and robustness.