Research On Animal Target Detection Algorithm Based On Convolutional Neural Network

The protection of biological resources is an important prerequisite for maintaining the balance of the ecosystem,and animal resources occupy an important position in biological resources.Protecting animal resources plays a crucial role in natural ecosystems.Humans' emphasis on animal resources is constantly increasing.In today's society,endangered wildlife requires regular human protection to prevent the loss of animal species diversity.In order to better protect animal resources and implement realtime detection of animals,this paper applies the target detection algorithm to animal images.The animal target detection method will classify and locate the animals appearing in the target image.The size of the animal target in the image is affected by the perspective,etc.The same species of animals may appear different in size,and the large and small distances also cause common large animals to appear in the picture at a small scale.In the process of detection,the accuracy of the detection algorithm will be reduced due to the influence of different factors on the size of the animal target.In order to solve the problem of animal target size,this article will study animal targets of different sizes.Aiming at the impact of animal size in the target picture,this article first introduces a variety of commonly used target detection algorithms,and then further studies Faster R-CNN and YOLOv3 two target detection algorithms.Animal images are used as research objects.Based on the algorithms of Faster R-CNN and YOLOv3,the two algorithms are improved and experimented.The main work of this paper is as follows:(1)An improved animal target detection algorithm based on Faster R-CNN model is studied.We use multi-scale convolution features and introduce multi-scale contextual information to resolve false detections and missed detections caused by animal targets in the image due to the small size.The method of high-level and lowlevel feature fusion can enhance the semantic information of low-level features and enhance the high-level feature details.At the same time,multi-scale context information processing is introduced,which indirectly helps identify animal target information in the picture,thereby improving the accuracy of animal target detection.(2)This paper studies an improved animal target detection algorithm based on YOLOv3 model.Aiming at the shortcomings in the detection rate of the two-step target detection algorithm,this chapter is based on the one-step target detection algorithm YOLOv3.Through cluster analysis,the prior frame of the animal target data set selected in this paper is optimized,and the multi-scale detection is improved.The size of the image and the addition of a new detection layer fuse different layer features to improve the accuracy of animal target detection.Under the premise of ensuring a high detection rate,improve the detection accuracy of animal targets.