Algae Object Detection Algorithm Based on Improved YOLOv5

Abstract

Algae are sensitive to changes in water quality, and the substances produced by algae also have a direct impact on the aquatic environment, so the detection and analysis of algae can provide important information on changes in water quality. In this paper, we propose an improved deep learning network for algae object detection, YOLO-MultiAlgae, which is capable of fast object detection of algae images. The method uses GhostNet as the backbone network and employs the SimAM attention mechanism to improve the detection accuracy. Experiments on the detection of eight species of algae show that the proposed network has higher labeling speed and accuracy compared to other classical object detection networks, with the detection accuracy reaching up to 98.8%, as well as better results in comparison experiments with two lightweight networks, shuffleNetV2 and MobileNetV3. Algae detection can be performed faster and better with this network.