基于暗通道先验与YOLO的水下河蟹识别研究

Recognition of Underwater Crab Based on Dark Channel Prior and YOLO

为解决全自动均匀投饵作业船在河蟹养殖过程中投饵不精准问题,引入水下摄像设备采集图像,但采集的图像存在对比度低、模糊和图像退化等问题。为此,采用暗通道先验提高图像对比度。利用YOLO卷积神经网络技术快速准确地识别出低照度环境下的河蟹,识别准确率达到98%,平均耗时50ms。获得河蟹生长、分布信息后测算出河蟹养殖密度,为全自动均匀投饵作业船精准投饲提供数据支持。

In order to solve the problem of inaccurate baiting of fully automatic uniform baiting boat in crab culture,this paper introduc⁃es underwater camera equipment to collect images.Due to the problems of low contrast,blurring and degradation in underwater crab pond image,this paper uses dark channel prior to improve image contrast,and YOLO convolution neural network technology to identi⁃fy crabs in low illumination environment quickly and accurately.The recognition accuracy rate reaches 98%,and the average time con⁃sumes 50 ms.This method can achieve fast and accurate identification of river crabs in complex underwater crab ponds,obtain the growth and distribution information of river crabs and estimate the culture density of river crabs,which lays the foundation for accurate feeding of fully automatic uniform baiting boats.