基于语义分割网络的克氏原鳌虾图像识别与分割

Recognition and Segmentation of Crayfish Images Based on Semantic Segmentation Network

为高效获取克氏原螯虾的精准分割图像,提出了一种基于语义分割网络的克氏原螯虾图像识别与分割的方法。首先,在克氏原螯虾上市季节(4—8月)拍摄其图像,建立数据集;然后使用Labelme对所采集的数据集制作标签,进行数据增强,并将数据集按8∶2的比例随机划分为训练集和验证集;最后基于U-Net语义分割网络基础来对数据集进行预处理操作,其中网络的编码器部分捕捉克氏原螯虾图片的上下文信息,解码器部分对克氏原螯虾图片进行精准定位,完成分割任务。结果表明,训练后网络模型的validation Dice(验证集相似度)达到0.944 5,validation loss为0.055 5。这表明该研究所提出的语义分割网络可以较好地解决克氏原螯虾识别与分割问题,能够实现准确、高效的克氏原螯虾图像分割,为实现克氏原螯虾机器化优质选种打下了的基础,也为水产养殖智能化提供了视觉支持。

To efficiently obtain accurate segmented images of crayfish(Procambarus clarkii),a semantic segmentation network based method for image segmentation and recognition of crayfish was proposed.Firstly,during the marketing season from April to August,we took photos for crayfish and established a dataset;then used Labelme to label the collected dataset,performed data augmentation,and randomly divided the dataset into training and validation sets in an 8:2 ratio;finally,based on the U-Net semantic segmentation network,the dataset was preprocessed:the network encoder captured the contextual information of the crayfish image,and the decoder accurately located the crayfish image to complete the segmentation task.The experimental results showed that the validation Dice(validation set similarity)of the trained network model reached 0.9445,and the validation loss was 0.0555.That means the semantic segmentation network proposed in this article can effectively solve the problem of identification and segmentation of crayfish,and achieve accurate and efficient image segmentation of crayfish,which may provide an important foundation for realizing mechanized selection of excellent crayfish resources and also offer visual support for intelligent aquaculture.