Research on Underwater Plastic Waste Detection Technology Based on Improved YOLOv8

In today 's era, plastic waste has become an important source of water environmental pollution. In order to reduce plastic waste pollution and protect the water environment, this study proposes a lightweight detection model based on improved YOLOv8 for underwater plastic waste detection. Firstly, the image in the public dataset is enhanced by using the composite image enhancement technology to improve the contrast and resolution of the image. Then, the YOLOv8 s model is improved: the non-parametric attention mechanism SimAM is introduced in the head part to highlight the key information of the target and suppress the irrelevant information, so as to improve the representation ability of the model. Deformable Convolutional Networks v3(DCNv3) is introduced to adapt to the deformation of the target and improve the detection accuracy of the model. The YOLOv8 backbone is replaced with a lightweight network MobileNetV3 to reduce the number of parameters and improve the model inference speed. The experimental results show that the recognition accuracy of the model for plastic waste reaches 84.9 %, which is 5.1% higher than the mAP50 before the improvement, and the total time of single image processing is 3.4 ms. The model realizes the detection of various underwater plastic waste with high detection speed and high detection accuracy. The model is of great significance for the detection of underwater plastic waste and the protection of water environment.