Object detection of inland waterway ships based on improved YOLOv7

In recent years, with the rapid development of deep learning, more and more deep learning technologies have been applied to the field of ship detection. Compared with traditional target detection algorithms, deep learning target detection algorithms are more robust, have stronger generalization ability, and are easier to be applied to actual scenarios. On the premise of summarizing existing ship detection algorithms and based on the YOLOv7 detection framework, this paper aims at the characteristics of small target and high density of inland waterway ships in this paper. By introducing the improved K-Means++ anchor frame reunion class, adding a fourth small target detection layer, CBAM attention mechanism, SIoU positioning Loss function and Varifocal Loss classification loss function, and combining and comparing each algorithm to select the most suitable combination algorithm to solve the problem of ship target detection in the actual scenario. The original YOLOv7 network and the improved YOLOv7 network were used for experimental comparison on the self-built data set of inland waterway ships. Compared with the original network, the missing phenomenon of the improved YOLOv7 network model was greatly reduced, and the mAP of the improved YOLOv7 network model reached 90.6%, which increased by 13.7% compared with the original network model. The detection effect is better than the original network.