Design And Implementation Of Animal Face And Key Point Detection Algorithm Based On Deep Learning

In the process of intelligent development of animal husbandry,accurate identification of individual animals is an indispensable part.And the facial recognition technology is a very important branch of individualized accurate recognition technology.In the face recognition technology,the face and its key point detection technology is also the difficulty in this field.Therefore,this paper design an animal face and key point detection algorithm with higher accuracy and faster detection speed based on deep learning technology.First of all,according to the characteristics of animal facial features,the traditional facial feature detection algorithm MTCNN face detection algorithm is used as the basic network framework for animal facial and key point feature detection.Then,according to the multi-scale attributes of animal facial features,a multi-scale feature extraction network was designed using convolutional neural networks to select candidate frames,and the candidate frames were normalized by RoI Align.The function of automatic extraction of candidate regions,and then complete the extraction of multi-scale features of animal facial features.It saves the time of MTCNN network feature extraction,and improves the detection accuracy and recall rate of the network model for smaller facial regions.In addition,according to the characteristics of a large amount of image data and animal facial features in the training process of the traditional model,a convolutional neural network is used to design an animal face and human face migration model,and the human face data is processed by affine transformation.Feature migration,using the face data after feature migration to pre-train the improved MTCNN network model,and then on the basis of the pre-trained network model,use the animal face data to perform Fine Tuning training on the model,and then realize the animal face and its key Accurate detection and positioning of points.Experimental results show that the use of multi-scale feature extraction to extract image candidate regions can improve the model's detection effect on small objects.Use face conversion model data to pre-train the network model,and then use Fine Tuning training to train the model.The effect of the network model is better.Compared with other detection methods,this method makes the network model have a higher accuracy and recall rate.Meet the expected design requirements of the model.