基于CNN和图像深度特征的雏鸡性别自动鉴别方法

Automatic Recognition Method of Chick Sex Based on Convolutional Neural Network and Image Depth Features

为有效辨别雏鸡性别,提高养鸡效益,针对部分雏鸡的泄殖腔特征不明显、采集雏鸡泄殖腔图像易受光线影响的问题,提出了一种基于卷积神经网络和图像深度特征的雏鸡性别自动鉴别方法。以翻肛法采集的雏鸡泄殖腔图像为研究对象,利用卷积神经网络构建待识别雏鸡泄殖腔的深度特征和雏鸡泄殖腔的深度特征向量集合库;将待识别雏鸡泄殖腔的深度特征与雏鸡泄殖腔的深度特征集合库进行相似度比较,并对比较结果进行排序;将排序结果中排在前n个与待识别雏鸡泄殖腔图像最接近的深度特征,与待识别雏鸡泄殖腔的深度特征进行特征融合,再通过卷积神经网络进行识别。结果表明,本文方法在测试数据集的识别准确率达到了97.04%,在生产环境下识别准确率达到了96.82%,相比常规的卷积神经网络方法,本文方法提高了雏鸡性别的识别准确率。

Aiming at the problems of some chicks’unobvious cloacal features and the influence of light on the collection of chicks’cloacal images,a method of automatic recognition of chick sex based on convolutional neural network(CNN)and image depth features was proposed to effectively distinguish male and female chicks and enhance the benefit of raising chickens.Taking chicks’cloacal images collected by the method of anal examination as the research object,a CNN was used to establish vector collection libraries,including the in-depth features of both chicks’cloacal images to be identified and chicks’cloacal images.Similarity comparison was performed between the collection libraries of the in-depth features of chicks’cloacal images to be identified,and chicks’cloacal images and the comparative results were ranked.Feature fusion was conducted for the in-depth features that were ranked top n in the ranking results and were the most similar to chicks’cloacal images to be identified and the in-depth features of chicks’cloacal images to be identified.The depth characteristics of the clonal cavity of the chick were highlighted,and then identification was carried out via CNN.The experiment results showed that the accuracy on the test dataset reached 97.04%,and in the production environment reached 96.82%.Compared with conventional CNN methods,it improved the recognition rate for identifying male and female chicks’cloaca.