Improved YOLOv8-based Algorithm for Maritime Vessel Object Detection
Tianrui Niu · 2024
In response to the challenges of low accuracy, significant errors, slow speed, and inconsistent dataset quality in maritime vessel object detection tasks, this paper proposes an improved algorithm based on the YOLOv8 framework. The objective is to enhance the accuracy, speed, and robustness of the detection algorithm while addressing the variations in dataset quality. Firstly, data augmentation is utilized as a preprocessing technique to increase the model's robustness and generalization capacity. By augmenting the training dataset with diverse maritime vessel images, the algorithm learns vessel-specific features and becomes more capable of handling variations in the data.Furthermore, a lightweight channel attention mechanism called Channel Attention (CA) is incorporated to strengthen the focus on inter-channel relationships and long-range positional information. This helps to reduce detection errors and improve the overall performance of the model.To improve the model's receptive field and feature representation capabilities, the SPPFCSPC module replaces the SPPF module. This modification enhances the model's ability to capture contextual information and improve the accuracy of vessel detection.Additionally, the traditional Confidence IoU (C IoU) metric is replaced with the Wise IoU, which reduces the competitiveness of high-quality anchor boxes and mitigates the harmful gradients generated from low-quality examples.Experimental results demonstrate that the improved algorithm achieves a 3.6% improvement in [email protected] on the validation set and a 1.4% improvement on the test set compared to YOLOv8. Moreover, the detection speed is increased by 10 frames.