Research And Application Of Object Detection Via Deep Visual Perceptive Learning

Object detection is a very important research direction in the field of computer vision.It is the basis for the computer to recognize and understand the image.It is also the premise of the computer to make judgments,inferences,and decisions.It can be widely used in enemy object detection of remote sensing image,object localization in robotics,and computer-aided diagnosis in the medical field and so on.The object localization in space aims to find the three-dimensional coordinates.The application of object detection in spatial localization provides a method from image processing in two-dimensional to the perception in three-dimensional space.So,it is very significant to research object algorithm and application.With the constant innovation of deep learning,the object detection still has problems in the development of leapfrog development.Such as the accuracy and efficiency are poor for the high-resolution image,the accuracy of object detection isn't high for the small samples,especially in complex scenes,and the application of object detection in object spatial localization is immature.Therefore,we focus on the above problems,further study the object detection algorithm and application based on deep visual perceptual learning*.(1)To improve the accuracy of detection for h igh-resolution images.An effective visual perceptive object detection approach based on YOLO(you only look once)is proposed in this paper.The proposed algorithm consists three modules mainly: subregion extraction based on visual attention,sub-region object detection,and semantic correlation suppression.Firstly,we can select some sub-regions in the scene using saliency maps.It can transfer computing resources to the area that may contain objects to reduce the computational complexity.Then,pre-selected objects are obtained by YOLO which is a fast learning model.The short-cut is used in the convolutional neural network,which can maintain the high real-time performance and improve the detection accuracy of the small object.Finally,we get final re sults by object semantic relevance suppresses which is proposed in this paper.It can reduce the interference of false objects for reducing the false alarm.Comparing with the classic algorithm,experimental results demonstrate that the performance of high-resolution image is improved by the proposed algorithm.(2)In the research process of(1),we find that the training set is the key factor to influence the accuracy of detection.This paper proposes an object detection algorithm based on deep supervised learning for small samples.Firstly,the algorithm can augment training samples automatically by synthetic samples generator.The synthetic samples generator is designed by switching the object regions in different scenes.Then,deep supervision learning and dense prediction structure are used in the deep convolution neural networks.It is a better solution to solve the vanishing-gradient and the objects with different scale.In addition,the semantic relevance of objects is used to improve the accuracy of weak-feature objects in complex scenarios.Experiments on B3 DO demonstrate that the proposed algorithm achieve s better results than the stateof-art contrast models.(3)To further improve the application of object detection in object spatial localization.We combine depth image and detected objects to measure this objects and locate this objects.Object detection provides the semantic information and the pixel localization,and the depth image provides the distance between the object and Kinect.The combination of the object detection results and the depth image will improve the accuracy of measurement and positioning.The application mainly includes four key steps: source data acquisition,object detection model integration,camera calibration,object dimension measurement,and spatial location.Experiments demonstrate that the accuracy of spatial localization and dimension measurement is high.This paper designs a software named microvision object perception and positioning,which can be applied to the non-contact object dimension measurement and space positioning in the environment such as high temperature or closed space.