Research On Action Recognition Of Dairy Goat Based On CNN

The dairy goat is a dairy breed goat with large milk yield and high economic returns.Its goat milk is comprehensive in nutrition and is an important raw material for the modern dairy industry.For the behavior recognition needs of large-scale breeding dairy goats,this study constructs and trains appropriate convolutional neural networks,uses Dense Net model,optimizes experimental parameters,and trains a stable convergent convolutional neural network for the collected dairy goats.Behavioral images are classified and identified.The data set is further classified and the behavioral image of the dairy goat with multiple tags can be trained through the improved SE-Dense Net model combined with the multi-label classification algorithm.Experiments show that the improved SE-Dense Net performance is better than the traditional Dense Net,and the multi-label classification of the new milk goat data set has been achieved,and a good recognition effect has been achieved.In this paper,the research object is set as the daily behavior image of the dairy goat collected by the Xiong dairy goat base,and the problem of milk goat behavior recognition is solved by convolutional neural network.(1)Image PreprocessingScreening dairy goat images with obvious behavioral characteristics and classifying them,manual deletion of blank images without milk goats,or less than two-thirds of the total body of dairy goats,preliminarily pre-processing the input image,unified pixels and do image enhancement.(2)Action recognition of dairy goat based on Dense NetThe Dense Net model is constructed.The focus of this paper is to construct a reasonable network model structure,and adopt relevant techniques to ensure that it converges quickly and stably on the training set,and optimizes the relevant parameters.The pre-processed image is input and the target features of the image are extracted to train the Dense Net model.Finally,the softmax classifier is used to classify the training target,and finally the result of the recognition test is obtained.Then,the test set image is input for testing,and finally the accuracy of the verification result is recognized.(3)Action recognition of dairy goat based on multi-label SE-Dense NetAn improved Dense Net model is proposed,which is added to the SE module in the SENET model.The improved SE-Dense Net model combines the advantages of both,strengthens the transmission of deep information,avoids the disappearance of the gradient,and realizes the feature extraction process.Features are recalibrated.At the same time,the behavior dataset of dairy goats was reclassified,and the network output was encoded by the heat.The sigmoid cross entropy loss function was used to realize the multi-label behavior of dairy goats.