Wildlife Monitoring Image Classification Based On Attention-CNN-GRU

As a non-intrusive monitoring method,infrared cameras are widely used in the field of wildlife monitoring.However,this method can hardly realize real-time monitoring of wildlife and it requires a amount of time to distinguish between falsely triggered images and wild animal images.In view of the above problems,this thesis proposes a new solution for wildlife monitoring and classification by combining wireless image sensor network and deep recurrent neural network.In terms of real-time transmission,this thesis realized the real-time collection,transmission and storage of wild animal monitoring images through the hardware system of wireless image sensor network.In terms of classification,this thesis realized efficient classification effect with the adoption of the Attention-CNN-GRU based wildlife identification method.Major work of this thesis is presented as follows1.A wildlife monitoring system based on wireless image sensor network was designed.Given the complex environment of the nature reserve and the dynamic environment,this paper analyzed the practical requirements of the wildlife monitoring system and proposes the systematic design.The monitoring system is designed from two aspects,namely software and hardware.Monitoring nodes,coordination nodes and server nodes cooperate to realize remote,real-time and autonomous collection and transmission of wildlife monitoring images2.An image data set of monitoring wildlife was established.On the basis of wireless image sensor network and infrared camera,this thesis sorts the images collected from the public dataset and the wild field.The established data set contains 33 species of 62130 wild animal images.By analyzing the correlation coefficient and cosine similarity among images,the temporal and spatial correlation of the captured images of the monitoring nodes is proven,which provides research materials and theoretical support for the classification and identification of wildlife.3.A method of wildlife identification based on Attention-CNN-GRU is proposed.On the basis of correlation among the wildlife images taken by the monitoring nodes,this thesis applies the Recurrent Neural Network into sequence analysis of images,which effectively extracts the time series information between wild animal pictures.A classification method of wildlife based on Attention-CNN-GRU was proposed by linking the correlation coefficient with the attention mechanism.The experimental results showed that the Attention-CNN-GRU-based wildlife monitoring and classification system can effectively classify false trigger and wild animal images generated by monitoring nodes.