Research On Bird Target Detection And Recognition Model Based On Convolutional Neural Network

Bird damage is widespread in various fields,and bird oriented target detection and recognition is of great significance.Traditional bird detection and identification rely on manual work,which requires experts to have better knowledge and experience reserves,with low stability and efficiency.With the continuous development of deep learning,target detection technology can be applied here to improve the efficiency of detection and recognition.Yolov3 shows an excellent level in this field.It has the characteristics of simple structure and good reproducibility.In this paper,the algorithm is used as the original model,and several improved strategies are used to improve the performance of the model,and a bird detection system is constructed.The main work is as follows:1)A bird detection model based on convolutional neural network is constructed.For the unstable features of different sizes in model detection,an adaptive feature fusion module is added.The structure can adaptively acquire the weight of each dimension feature,filter out interference elements,and enhance the reliability of feature information;The k-means++algorithm is used instead of the K-means clustering algorithm to optimize the steps of generating the starting point,which is conducive to the generation of the anchor box.Compared with the original model,the map value of the model with ASFF structure and k-means++ is increased by 1.84%.2)Bird detection model based on improved loss function.Ciou is used to replace the MSE as the loss function of the model.This function can comprehensively consider the coverage area,center point spacing and aspect ratio,finally more accurately reflect the detection results,which is conducive to more stable regression and has certain robustness;Spp architecture is introduced,which can transfer the obtained features of multiple sizes to the channel fusion of the feature graph,and enhance the ability of the algorithm to detect targets of different sizes.Compared with the original model,the map value of the improved loss function and the new spp structure bird detection model increased by 4.6%.3)The bird detection system is established on the web through Python’s flask architecture.From the perspective of mobile terminal,the user requirements and resource allocation in Android system are analyzed.The improved bird detection model is deployed to the mobile terminal to complete a lightweight bird detection system based on mobile devices,which can meet the needs of portable bird detection...