Research On Dim And Small Target Detection In Infrared Images

Object detection is one of major directions in the computer vision research field.It has important research significance and broad application prospects in image retrieval and video surveillance.In simple terms,object detection is to analyze the information contained in the image itself to obtain the location and category information of the object of interest,and finally it can achieve intelligent analysis of the picture scene.With the rapid development of information technology,the development of object detection technology is also changing with each passing day.And in the filed of object detection,the detection of dim small targets in infrared images has high value in both military and civilian fields.This thesis mainly studies the method of dim small targets detection in infrared image,and it mainly uses drones captured by infrared cameras as targets to be detected,it tests and analyzes the tranditional theory of infrared dim small target detection,and improves the target detection method based on deep learning to make it enable for infrared dim small target detection.The main contents of this thesis are as follows:1.The traditional theories of infrared dim small target detection are introduced,and some representative methods for each theory are selected to carry out simulation experiments.Representative target detection methods based on deep learning are introduced and their detection capabilities for infrared dim small targets are verified.It is proved that there are lots of deficiencies to use these target detection methods based on deep learning to train and detect weak infrared targets,and these mathods need to be improved.2.The main idea of target detection based on FreeAnchor theory is introduced,and the reason for the high recall rate of this detection method is explained.Based on the FreeAnchor detection method,its backbone and normalization mode are imporved to make it can extract better features,and enable it to detect small infrared targets.The performance of the method is evaluated,and it is proved that the method can detect infrared dim small targets,and rise the detection rate.3.The main idea and algorithm structure of YOLOv3 target detection are introduced.Based on the YOLOv3 target detection method,an improvement in fast detection of dim small targets in infrared image is made.The detection method`s backbone,the way to extract features,loss function and activation function are optimized.The performance of the method is evaluated,and the experiment results show that the method can achieve the accurate detectionof dim small targets in infrared images with a little loss of real-time capability.