穆春华1,2，范良忠2，刘 鹰3
（1太原科技大学电子信息工程学院，太原 030024；
2浙江大学宁波理工学院信息与工程学院，宁波 315100；
3 中国科学院海洋研究所，青岛 266071）

摘要：利用计算机视觉技术和机器学习方法研究工业化循环水养殖的残饵与粪便的识别问题，为基于残饵浓度检测的智能投喂系统提供理论依据。首先对残饵视频进行图像预处理，分割出残饵和粪便图像；然后根据残饵和粪便在灰度分布和形状上的差异，提取平均灰度，周长平方面积比、凸壳面积比、骨架数、对比度、逆差距6个特征；再分别运用4种不同核函数的SVM算法和改进的决策树算法进行残饵图像识别。结果显示，径向基核函数的SVM算法识别效果最好，残饵和粪便识别率分别达到99%和97%以上；改进离散方式的决策树算法识别率与SVM算法的识别率接近，并且实时性更好。

关键词：工业化循环水养殖；计算机视觉；决策树；支持向量机；残饵识别

Research on the residual feeds recognition of recirculating aquaculture systems based on computer vision
MU Chunhua1,2, FAN Liangzhong2, LIU Ying3
(1.School of Electronic Information Engineering, TaiYuan University of Science and Technology, TaiYuan, 030024, China;
2 College of information science and Engineering, Ningbo Institute of Technology, Zhejiang University, Ningbo,315100, China;
3 Institute of Oceanology, China Academy of Sciences, Qingdao, 266071, China)

Abstract: The paper mainly researches the residual feeds and feces recognition of recirculating aquaculture systems by using the computer vision technology and the machine learning methods, which provides a theory basis for intelligent feeding system based on residual feeds concentration detection. At first, the residual feeds and impurity images were obtained by preprocessing the residual feeds video record. Then, the features were extracted by analyzing the difference between residual feeds and feces in gray level distribution and shape. The features include: AverPixel, Peri2Area, Conv2Area, Skeletons, Contrast and IDM. Finally, we realized the recognition of residual feeds image by using the support vector machine algorithm based on 4 different kinds of kernel function and the modified decision tree algorithm. Experimental results showed the SVM based on the radial basis kernel obtained the best recognition rate. And the recognition rate of residual feeds and feces were up to 99% and 97% respectively. The recognition rate of decision tree with modified discrete way is close to the SVM’s, and the real-time performance is better.

Key words: industrial recirculating aquaculture；computer vision；decision tree；support vector machine；recognition of residual feeds