Target Recognition Based On Semantic Structure And Visual Focus In Complex Scene

Object detection and recognition in complex scenes is always a hot research topic in the field of machine vision, artificial intelligence, pattern recognition and image processing,and is widely used in military science and technology, industrial security, daily life, medical care, weather reporting and other aspects. Just like Google and Baidu’s unmanned vehicle which is developping very rapidly in current,DJI-Innovati-ons’ s Phantom series,Iqiyi video advertising, Smart city of GoSunCn, as well as Magic Leap technology and so on, there is no place to see the shadow of the target detection and recognition. Therefore, it is very important to improve the recognition rate of targets in complex scenes and real-time performance.The improvement work which has been done in this paper mainly included the following two aspects.Firstly,the classical ITTI model combined with Pulse Coupled Neural Network(PCNN) can determine the objects of the whole scene,but the separation of the target objects will also separate into the background image, and also can not separate the individual goals out. Therefore, it will have a greater impact on the feature vectors to be extracted later, which has a serious impact on the recognition rate. The improved MeanShift+ITTI model achieve a good separation of the object,this method can separate the whole object from the scene,and only with a small amount of background information, but it is the same that this method can not achieve the separation of multiple target in one scene. Therefore, this paper proposes an improved ITTI model based on visual focus,which can achieve the separation of objects in multi object complex scenes, at the same time, it also has a better recognition effect on the situation of occlusion and adhesion between objects and objects.Secondly, the semantic mechanism of natural scene from natural language processing is proposed in this paper, which is applied to the model of SVM classifier, for the scores of SVM classification which does not meet the natural scene semantic mechanism, select the next highest score as the judgement by category, and so on, so as to make the result of target recognition meet the natural scene semantic mechanism, and only a slight delay in time, also the recognition rate has been improved.In summary, the object detection and recognition technology based on semantic structure and visual focus which proposed in this paper, has the characteristics of good adaptability, high recognition rate and good real-time performance.It can be applied and referenced in a variety of target recognition technology.