基于改进YOLO卷积神经网络的水下海参检测

Underwater sea cucumber identification based on improved YOLO convolutional neural network

为了实现水下海参的自动化捕捞,需要利用机器视觉方法实现水下海参的实时检测与定位。本研究提出一种基于改进YOLOv5s的水下海参检测定位方法。针对海参与水下环境对比度较低的问题,引入多尺度视觉恢复算法对图像进行处理,增强图像对比度;为了提高模型特征提取能力,加入了注意力机制模块;原始模型对YOLOv5s小目标的检测效果不佳,改进后的YOLOv5s模型替换了原有的激活函数,并在Head网络中加入了新的针对小目标的Detect层。使用改进的YOLOv5s模型与YOLOv5s、YOLOv4和Faster-RCNN在相同的图像数据集上进行试验,结果表明,改进的YOLOv5s模型的检测精度和置信度,尤其是对小目标的检测效果优于其他模型。与YOLOv5s模型相比,改进后的YOLOv5s模型的精度和召回率分别提高了9.6个百分点和12.4个百分点,能够满足水下海参的实时检测要求。

To realize automatic fishing of sea cucumber underwater,it is necessary to use machine vision method to realize real-time detection and positioning of underwater sea cucumber.In this study,a detection and localization method for underwater sea cucumber based on improved YOLOv5s was proposed.Aiming at the problem of low contrast between the sea cucumber and the underwater environment,a multi-scale vision restoration algorithm was introduced to process the images to enhance the contrast of the images.The attention mechanism module was added to improve the feature extraction ability of the model.The original model didn’t show good detection effect on small object of YOLOv5s.The improved YOLOv5s model replaced the original activation function and added a new Detect layer into the Head network which aimed at small object.The improved YOLOv5s model was used to conduct experiments with YOLOv5s,YOLOv4 and Faster-RCNN on the same image data set.The results showed that,the improved YOLOv5s model showed better detection accuracy and degree of confidence compared with other models,especially for small target detection.Compared with the YOLOv5s model,the precision and recall rate of the improved YOLOv5s model increased by 9.6 percentage points and 12.4 percentage points respectively,which could meet the requirement of real-time detection of underwater sea cucumber.