Abstract:

In order to solve the problem of intelligent baiting in the process of outdoor pond culture， an intelligent control method of baiting system based on machine vision was proposed. 4 texture feature parameters of entropy， energy， correlation and contrast were extracted from the gray-scale co-occurrence matrix to quantitatively evaluate the feeding intensity of the fish. According to the statistical results， the feeding intensity of the fish was divided into 3 types： no feeding， weak feeding and strong feeding. The independent weight method was used to determine the best weight of the 4 parameters and establish a parameter weighted fusion model. The results showed that the weight of entropy was the highest， reaching to 47.04%， the weight of correlation was the lowest with 12.70%， and the weights of energy and contrast were 27.55%， 12.71%； using support vector machine to identify the 3 feeding inteisities of fish， the recognition accuracy reached 99.77%， and the processing time of a single frame image was 0.48 s. Above method was simple and fast， and could meet the technical requirements of the intelligent control of the feeding system， and provided a new idea for the intelligent control of the feeding system for outdoor pond culture.

Key words: outdoor pond farming, machine vision, fish feeding status recognition, texture feature, intelligent feeding

摘要：

为解决室外池塘养殖过程中的智能化投饵难题，提出一种基于机器视觉的投饵系统智能控制方法。利用灰度共生矩阵提取熵、能量、相关性、对比度4个纹理特征参数对鱼群摄食强度进行量化评估，根据统计结果将鱼群摄食状态分为未摄食、弱摄食、强摄食3种状态；采用独立性权重法确定4个参数的最佳权重并建立参数加权融合模型。结果表明，熵的权重最高，达到47.04%，相关性的权重最低，为12.70%，能量与对比度的权重分别为27.55%、12.71%；利用支持向量机对鱼群的3种摄食状态进行识别，识别准确率达到99.77%，单帧图像处理时间为0.48 s。以上方法简便快速，能够满足投饵系统智能控制的技术需求，为室外池塘养殖投饵系统的智能控制提供了一种全新思路。

关键词: 室外池塘养殖, 机器视觉, 鱼群摄食状态识别, 纹理特征, 智能投饵