SAR Image Target Detection Technology Based On Environmental Cognition

Target detection technology is to extract the location of the instance target from the specified image or video and identify it.It is an important research direction in the field of computer vision at the same time,and it is also used in many fields such as pedestrian detection,vehicle monitoring,and face recognition..At present,SAR image target detection based on deep learning has the problems that the data set is difficult to obtain and mark,and at the same time,it does not use the environmental information around the target well.Therefore,the main research goal of this paper is to use the method of simulating and synthesizing data sets,and propose a target detection model that can effectively use environmental information,effectively solve the above problems and improve the detection accuracy of the model.First,for the problem that SAR images are difficult to obtain and have a certain degree of difficulty in interpretation,this paper uses the region growth algorithm to segment the sliced targets in the MSTAR dataset,extract the main targets and shadow areas,and then replace them according to certain distribution rules.Pixels in the target scene complete the construction of the data set with the target scene.On this basis,considering the characteristics of the SAR image,the size of the target in the series of images changes little,the size of the candidate frame in the target detection model based on the candidate frame can be fixed,and the corresponding improvement of the model is verified by experiments This operation can effectively reduce the network test time.Then,in view of the problem that there are many false detection results in the target detection network results with fixed anchor points,this paper analyzes the elements that are easy to misjudge in the environment,and uses the statistical information of the image in turn,the information processing model based on machine learning,and the deep learning The process of the information processing model performs post-processing on the verification result set of the target detection model fixed by the anchor point to test the performance of each method.The experiment proves that the simple conv-structure can better reduce the number of false detection results of the target detection model.