The Application Research Of Small Target Detection Based On An Improved YOLO

Target detection is to detect all objects in the image or video through the computer network,in order to detect the required target objects.Target detection plays an important role in the current machine vision,because the purpose is to find the target object,so it has a wide range of applications in public security,monitoring facilities system,public security management,agricultural fruit picking and other fields.Yolo(you look only once)algorithm has a good advantage in detection speed,but in small target detection,the convolution part of feature extraction is weak,and there are some disadvantages in the selection of candidate boxes.In this paper,Yolo algorithm is improved in two aspects: fruit detection and security X-ray machine image detection: in fruit detection,due to the observation distance,occlusion and light influence,it is easy to cause missing detection.In this paper,we improve the convolution kernel and use the adaptive non maximum suppression algorithm to reduce the detection of the fruit.In the security detection,the current two-step algorithm is slow,in addition,it will cause the problem of false detection and inaccurate target positioning.In this paper,the improved convolution kernel,dense convolution block and adaptive non maximum suppression algorithm are used to improve.The innovation of this paper is as follows:(1)build data set by myself(2)two kinds of convolution kernel improvements are made for the small target situation (3)for the gradient problem of the network,the dense convolution block is used to replace the partial residual block(4) for the problem of occlusion missing detection,an adaptive non maximum suppression algorithm is proposed.Based on the fast detection characteristics of Yolo algorithm,this paper improves the network for two kinds of small target detection problems: fruit detection and security detection.In the fruit detection task,the recall rate is increased 3.72% In the security inspection,the recall rate is increased 2.27%,up to 98.55%.