Research On Object Detection Algorithm Based On Convolution Neural Network

With the rapid development of urbanization in our country,the intelligent monitoring technology has been applied to the construction of the city.The core technology of intelligent monitoring is the object detection.However,there are still some unsolved problems in object detection.For example,using the artificially designed features,duplicate detection and error detection and so on.So there is important significance to keep on researching object detection.Convolutional neural networks are widely used in visual tasks and get the best performance based on the ability of active design and extraction of image features.So this paper applies convolution neural networks to the object detection task,replace the image features of active designed,improve the detection accuracy.In addition,obtain adaptive non-maximal suppression algorithms by mathematical modeling,reduce duplicate detection and error detection,improve the detection accuracy.First of all,not only the application background of the object detection,but also the domestic and foreign researches development of this subject are introduced.For example,the background,typical model,basic principle of deep neural networks,the function and calculation method of each layer,forward propagation and back propagation,connection mode and so on.Provide a theoretical basis for the improvement of object detection algorithm.Additionally,this paper research the existence of traditional object detection algorithm.On the basis of it,the object detection based on deep features of this paper is put forward.In this paper,active design features replaced by the deep features extracted by convolution neural networks,due to the artificial design features be used in traditional object detection algorithm and with the characteristics of complex computing.Resolves problem of complex calculations and accuracy is not high.Finally,achieved automatic design and extraction ofimage features,improved the accuracy and the robustness of object detection effectively.Finally,this paper research the existence of non-maxima suppression algorithm of fixed threshold.On the basis of it,the non-maxima suppression algorithm based on adaptive threshold of this paper is put forward.Due to the area overlap ratio is fixed of non-maxima suppression algorithm,and with duplicate detection and error detection,this paper propose achieve adaptive area overlap rate calculation by statistics the means and variance and establish a standard area overlap rate threshold calculation model.The improved algorithm can suppress duplicate detection and error detection,improves generality of non-maxima suppression algorithm,improved the accuracy of object detection algorithm.