Joint Multi-attention Cascaded LSTM Model For Pig Face Expression Recognition

According to zoologists,animals express their emotions to the outside world through facial expressions.Facial expressions are important information that conveys their emotion.It is a comprehensive reflection of the physiology,psychology and behavior of livestock,and can be used to evaluate animal welfare.Therefore,the successful recognition of animal expressions is of great significance in the field of animal welfare.After understanding the emotions expressed by animal,it can effectively reduce animal injuries,ensure the freedom of animal expression,reduce animal fear and anxiety,and make animals live in a state of no hunger,thirst and worry.Our paper takes the pig as the research object and uses the deep learning method to identify and verify the pig face expression database established in this paper.Due to the simple structure of animals’ facial muscle,it is difficult to recognize the subtle changes in animals’ facial.In this paper,by adding a multi-attention mechanism,based on the convolutional network,the key area features of the pig face are extracted,and then the feature fusion is input into the long and short-term memory network in the form of frame sequence for facial expression recognition.In response to the above problems,this article proposes a joint multi-attention mechanism Cascaded LSTM model framework to classify and recognize pig’s temporal facial expressions for the first time.The specific research work is as follows:(1)First of all,the data set is expanded.There is currently no public animal facial expression data set.This article uses the data set to authorize shared data set resources.The original data set is manually classified into four types of expression samples of happy,anger,fear,and neutral.Each type of sample contains 80 expression short video segments,and the facial area in the data set is marked,a total of 320 video samples.In this work,the original data set is expanded by data enhancement processing methods to obtain 2240 video segment samples.(2)Secondly,this paper proposes a simplified multi-task cascaded convolutional neural network(SMTCNN),with the help of the SMTCNN to locate the pig face,extract the pig face part and normalize it.Then,the expression videos are passed into the multi-attention convolutional network(MACNN)and the long short-term memory network(LSTM)for feature extraction and expression recognition.The proposed method can effectively extract the key area features of the pig face required by the network according to the difference of animal expressions,and at the same time ensure the timing of pig face expression recognition.(3)Finally,the identification method proposed in this paper is validated by a comparative experiment.The comparative experiment is divided into the following two parts: First,the validity of the multi-attention mechanism is verified.The experiment shows that adding the multi-attention mechanism is better than using only convolution.The recognition accuracy of facial features extracted by the network is 6.304% higher on average;second,the traditional classic facial expression recognition algorithm is migrated to pig face expression to compare the methods proposed in this paper.The experimental results show that the algorithm proposed in this paper is in the same environment.It has higher recognition accuracy,higher practicability and better effect.