Research On The Technology Of Image Object Detection Based On Convolutional Neural Network

Image object detection is one of the most important and challenging problems in the field of computer vision.Image feature extraction is the key to image object detection.Convolutional Neural Network(CNN)can automatically learn the features needed for object detection from a large amount of data.So there is no need to manually design features,which greatly reduces labor costs.The object detection algorithm based on deep learning is robust to the scaling,deformation and changes of the surrounding environment of the detected object.And a deep learning network can detect many types of objects.Therefore,object detection based on deep learning is receiving more and more attention.However,CNN still have difficulty detecting multi-scale targets,especially small targets.We propose improvements to these CNNs,and these improvements are applied to aircraft inspection,which shows that the research work in this paper has very important theoretical research significance and engineering application value.This paper studies the object detection technology based on deep learning,and makes some improvement to the Two-stage's object detector to improve their detection accuracy for image objects.The main work and achievements of this paper are as follows:1,Convolutional neural networks are not very effective for detecting multi-scale objects,especially small objects.This paper merges the pyramid feature of convolutional neural networks to get a new feature layer.Object classification and regression are performed on this new feature layer,which makes the algorithm more adaptable to multi-scale object detection.2,Aiming at the problem that the object proposals' feature of the traditional convolutional neural network is too simple.This paper designed a new feature extraction module extract better features of these proposals,and put it after the semantic pyramid feature.And this can make the classification and regression more accurate,and improve the accuracy of object detection.3,There is a certain gap between the region proposals which are given by the Region Proposal Network(RPN)and the ground truth.This makes the feature obtained by Ro I Pooling is only a part of the feature of the object itself.This brings certain difficulties to the classification and regression of the region proposals.This paper proposes an object detection method based on Maxout R-CNN.Maxout R-CNN can enrich the characteristics of the object proposals.And it facilitates subsequent classification and regression of the candidate region.Experimental results show that Maxout R-CNN can improve the performance of the object detection algorithm.4,We applied the improved convolutional neural network to detect the object in Remote Sensing Image.Taking the aircraft detection as an example,we illustrate the problems that need to be paid attention to when using the convolutional neural network to detect object in Remote Sensing Image.The effectiveness of the improved convolutional neural network for remote sensing image detection is verified by experiments.