基于改进Cascade R-CNN和图像增强的夜晚鱼类检测

针对光照不均、噪声大、拍摄质量不高的夜晚水下环境，为实现夜晚水下图像中鱼类目标的快速检测，利用计算机视觉技术，提出了一种基于改进Cascade R-CNN算法和具有色彩保护的MSRCP（Multi-scale Retinex with color restoration）图像增强算法的夜晚水下鱼类目标检测方法。首先针对夜晚水下环境的视频数据，根据时间间隔，截取出相应的夜晚水下鱼类图像，对截取的原始图像进行MSRCP图像增强。然后采用DetNASNet主干网络进行网络训练和水下鱼类特征信息的提取，将提取出的特征信息输入到Cascade R-CNN模型中，并使用Soft-NMS候选框优化算法对其中的RPN网络进行优化，最后对夜晚水下鱼类目标进行检测。实验结果表明，该方法解决了夜晚水下环境中的图像降质、鱼类目标重叠检测问题，实现了对夜晚水下鱼类目标的快速检测，对夜晚水下鱼类图像目标检测的查准率达到95.81%，比Cascade R-CNN方法提高了11.57个百分点。

Object Detection of Underwater Fish at Night Based on Improved Cascade R-CNN and Image Enhancement

The underwater environment at night has the characteristics of uneven illumination, big noise and low quality of underwater fish video. In view of this, in order to realize the rapid detection of fish targets in underwater images at night, a method of underwater fish object detection at night based on improved Cascade R-CNN algorithm and MSRCP image enhancement algorithm with color protection was proposed by using computer vision technology. Firstly, using the video data of the underwater environment at night, corresponding underwater fish images at night were extracted according to the time interval. The original extracted image was enhanced by MSRCP. Then the DetNASNet backbone network was used to train network model and extract underwater fish feature information. The extracted feature information was input into the Cascade R-CNN model, and the Soft-NMS candidate box optimization algorithm was used to optimize the RPN network. Finally, the underwater fish target at night was detected. The experimental results showed that the method can solve the problems of image degradation and fish object overlapping detection in the underwater environment at night, and realize the rapid detection of underwater fish target at night. Using this method, the accuracy rate of object detection with underwater fish image at night can reach 95.81%, which was 11.57 percentage points higher than that of the traditional Cascade R-CNN method.