小龙虾识别与度量关键技术研究与应用

Research and Application of Key Technologies for Crayfish Identification and Measurement

小龙虾的表型性状是水产养殖和遗传育种中非常重要的经济依据,为了精准实现对小龙虾的视觉检测与定位,并对其主体躯干表型特征进行测量。首先提出了一种基于Mask RCNN实例分割模型检测小龙虾的方法,不仅能快速识别小龙虾,还能对其进行实例分割,生成高质量的二值掩膜图。然后在此基础上,提出一种新的小龙虾度量算法:通过提取小龙虾的轮廓曲线,获取其中轴线,采用分区法提取小龙虾的特征区域,再利用凸包和凸包缺陷算法获取小龙虾的特征点,实现对其全长、体长、头胸甲、腹部、尾扇等主要部位的无接触度量计算。结果表明:Mask RCNN模型在小龙虾数据集上具有很好的泛化性能,模型分割准确率达到了94.6%,目标识别的平均检测精度达到了98.7%。各测量指标的平均绝对误差均在5 mm之内,比人工手动测量难度大大减小,并且稳定高效,重复性更好。该方法将有利于生产和选育过程中小龙虾的鉴别及结构尺寸的快速获取。

The phenotypic traits of crayfish provide an important economic basis in aquaculture and genetic breeding.In order to accurately realize the visual detection and positioning of crayfish and measure the phenotypic characteristics of its main body,this paper first proposed a method for detecting crayfish based on the Mask RCNN instance segmentation model,which could not only quickly identify crayfish but also perform instance segmentation on it,so as to generate high-quality binary mask maps.On this basis,a new crayfish measurement algorithm was proposed.Specifically,by extracting the contour curve of crayfish,the central axis was obtained,and the partition method was used to extract the characteristic area of crayfish.Then,the convex hull and convex hull defect algorithms were utilized to obtain the feature points of the crayfish,thus realizing the contactless measurement calculation of the crayfish's main parts such as its total length,body length,carapace,abdomen,and tail fan.The results show that the Mask RCNN model has good generalization performance on the crayfish dataset,and the accuracy of model segmentation reaches 94.6%.The average detection accuracy of target recognition reaches 98.7%,and the average absolute error of each measurement index is within 5 mm.The Mask RCNN model is easier than manual measurement,and it is stable and efficient,with better repeatability.This method is beneficial to the identification and rapid acquisition of the structure size of crayfish in the process of production and breeding.