基于目标深度分析的鱼体生长在线测量系统

针对我国水产养殖中无法实时在线无接触的自动估测鱼体重量的难题,利用水下摄像头获取鱼体的双目图像,针对水下图像的复杂性进行优化,通过深度学习网络YOLACT对图像进行实时分割,通过自定义约束条件筛选的确信特征点完成左右图像匹配,最后计算出鱼体深度,从而复原实际鱼体大小。并对鱼体在水下不同深度进行了测量准确性测试,结果表明在70cm处存在测量最优值,长度和宽度误差分别为6.5%和5.75%。本研究为实现水产养殖中的个体实时无接触生长监测,从而形成闭环的物联网化生产模型提供了一种可行方案。

Online measurement system of fish growth based on target depth analysis

Aiming at the problem that it is impossible to automatically estimate the weight of fish in real-time, online and contactless in China aquaculture. Binocular image of fish body obtained by underwater camera.Using the underwater camera to obtain the binocular image of the fish body to optimize the complexity of the underwater image.Real time image segmentation through deep learning network YOLACT,the left and right image matching is completed by the confident feature points filtered by user-defined constraints,the depth of the fish is calculated to restore the actual size of the fish.The results show that there is an optimal value at 70cm, and the length and width errors are 6.5% and 5.75% respectively.This study provides a feasible scheme to realize real-time contactless growth monitoring of individuals in aquaculture and form a closed-loop Internet of things production model.