Bird Identification In East Dongting Lake Based On Deep Learning

Dongting wetland plays an important role in breeding and overwintering birds in China.At present,the methods of "long-term squatting,hidden observation and regular nest inspection" are mostly used for bird monitoring.These traditional monitoring methods based on artificial are time-consuming,laborious and inefficient,which will also have a certain impact on the reproduction and survival of birds.Therefore,this paper uses convolutional neural network in deep learning technology to identify birds in East Dongting Lake,improve the efficiency of monitoring and identification,and provide AI technical support for Bird Monitoring and identification in nature reserve.The main research contents of this paper are as follows.(1)The bird data set of East Dongting Lake was manually constructed by crawler technology.The bird data set was cleaned,and the images such as blur,large difference and data expansion were eliminated.The bird data set of East Dongting Lake with 360 species and72,000 bird images was constructed.(2)Based on the bcnn model,the Res Net-BCNN network model is proposed.The residual neural network resnet-34 is selected for feature extraction.After bilinear feature pooling,a joint pooling layer is added to reduce the feature dimension.The method of joint pooling is proposed and a reasonable loss function is designed.It is verified that the accuracy of Res NetBCNN model is better than the original model and other models.(3)Aiming at the problems of small inter class differences and large intra class differences in bird images,a bird recognition method based on improved efficientnet model is proposed in this paper.Taking into account that the MBConv model uses too many subtle parameters to simplify the calculation of the original model,the number of channels in the model can be reduced,and the correlation of the original model can be reduced.At the same time,Adam optimization algorithm is used to accelerate the training convergence speed of the model and prevent over fitting.Finally,through experimental comparison and analysis,it is proved that the improved efficientnet algorithm proposed in this paper not only has high accuracy,but also occupies relatively less memory.(4)The experimental comparison between Improved Efficient Net and Res Net-BCNN verifies that Improved Efficient Net algorithm has higher recognition accuracy in the process of bird image recognition.Based on this,this paper designs and builds the bird recognition system of East Dongting Lake,and realizes the function of bird recognition of East Dongting Lake.This paper adopts the method of migration learning to train the improved model.The experimental results show that the model in this paper achieves better accuracy,and the performance is slightly improved compared with the previous model;Finally,a Dongting lake bird recognition system is designed and implemented,and the model is applied to practice.