Application Research Of Image Recognition Based On Deep Learning

Facial expression recognition and vehicle attributes recognition are two important applications of image recognition,the main process of facial expression recognition is very similar to face recognition,but has its particularity,its particularity lies in judging a facial expression is mainly based on the eye and mouth muscle movement.This paper proposes a facial expression recognition algorithm based on deep learning to enhance the expression of key region features,the goal is to improve the accuracy of facial expression recognition.Considering the problems of external factors such as occlusion,illumination,camera shooting Angle,etc.,a multi-label data set of vehicle attributes shot by multi-angle cameras under surveillance is created,and a vehicle attributes recognition algorithm based on multi-task convolutional neural network is proposed to improve the recognition speed and accuracy.The main work of this paper is as follows:(1)A facial expression recognition algorithm based on deep learning to enhance the representation intensity of key region features is proposed.It needs to obtain the location information of two key regions of each face image in advance,extract the feature maps of the whole face image and features of the two key areas are extracted from the feature maps.In addition,continue to extract features from the feature maps.Finally,multiple types of features are fused to strengthen the expression ability of the key areas' features.Several experiments have proved that our method can improve the accuracy by 3-5% after strengthening the expression of key regional feature representation for the same network.Finally,the best effect obtained in the fer2013 data set in this paper is 75.71%,0.51% higher than the accuracy of the current best method.(2)Considering the features of shallow depth of convolution neural network have more detail characteristics,while the deeper features have more global characteristics,a facial expression recognition algorithm based on multi-scale convolutional neural network to enhance the expression of key region features is proposed.The feature maps of shallow and deep layers are extracted,and the corresponding key regional features are extracted from the feature maps,the shallow layer and the deep layer are one supervision,and finally the shallow layer and the deep layer features are fused as another supervision.Considering the network complexity,a variety of special connection mode be joined to enhance the network training speed.(3)A multi-label vehicle attributes data set be created.Considering the connection between two attributes of vehicles,a multi-task network for vehicle attributes recognition algorithm is proposed.Multiple experiments have proved that our method is 98.04%,0.09% higher than the current best method in vehicle type recognition rate,the car color is 96.74% and 2.27% higher than the current best method in vehicle color recognition rate,and reduces the time consumption.