Goat Face Recognition With Fusion Of Deep Learning And Wavelet Features

Large-scale breeding is the main development direction of modern dairy goat industry,in which accurate and rapid individual recognition method of dairy goats plays a key role.In the traditional breeding process,dairy goats are worn with ear tags for individual recognition.However,it can cause pain and physical damage to goats,as well as trigger stress reactions or diseases.In addition,the use of ear tag is less efficient,raising the cost of hardware and labor,and can not efficiently prevent fraud in agricultural insurance.In recent years,with the popularization of livestock farm camera equipment and the development of artificial intelligence,livestock individual recognition method based on deep learning and computer vision technology has become an important research direction.This paper proposed a goat face detection model and a recognition model using deep learning features and wavelet features of goat face image,which provides a high-precision,non-contact and low-cost way for goat face recognition.The main work and conclusions are as follows:(1)Construction of goat face image data set.In this paper,images of Xinong Saanen dairy goats under different illumination and different angles were collected.In order to improve the quality of the data set,unqualified images were removed,and images with too high similarity were removed based on SSIM.After that,label Img annotation tool was used to label the goat face area,and label files in PASCAL VOC format were made to form the goat face detection data set.Finally,image cropping,scaling and data augmentation were carried out to achieve the expansion of sample size,forming a goat face recognition data set containing a total of 17 600 goat face images.(2)Lightweight goat face detection model based on improved YOLOv7.In order to achieve accurate and rapid detection of goat face,this paper used Ghost module to replace conventional convolution operations,added ECA lightweight attention mechanism,and optimized the positioning loss function of the model to SIo U Loss based on YOLOv7 model.The experimental results showed that the improved YOLOv7-Ghost-ECA-SIo U model effectively reduced the model size.Besides,Precision,Recall and m AP of the improved model were respectively increased by 3.80,2.65 and 3.15 percentage points compared with YOLOv7,which can effectively complete the goat face detection task.(3)Goat face recognition model based on wavelet transform and convolutional neural networks.To recognize goat individuals under farm conditions,a goat face feature extraction module based on 2D-DWT and convolution operation was designed in this paper to carry out feature fusion of deep learning features and wavelet features.Based on this module,the convolutional neural network was built to form the goat face recognition model DWT-Goat Net.The experimental results showed that the accuracy of the proposed goat face recognition model can reach 99.74% and 99.89%,respectively,on test set under different light conditions of daytime and night.The structure of the model is simple and the convergence is rapid.The feature extraction ability of the model has been improved after the introduction of wavelet features.Moreover,compared with some classical convolutional neural networks,the proposed model is more superior in the task of goat face recognition.The proposed model provides an effective solution for problems in the fields of precision breeding,agricultural insurance and animal welfare.