Research On Vehicle Recognition Of Grain Depot Based On Image Processing

Grain depot is an important base for grain storage.At present,the security monitoring system widely used in the smart grain depot has no identification function.It is difficult to effectively monitor the cheating of grain transport vehicles in grain depot,such as "changing cars and returning skins","inspecting A and selling B",which not only need professional managers to monitor at all times,waste a lot of human resources,but also the grain property in our country can not be effectively guaranteed.Therefore,it is urgent to add the identification function of grain depot vehicles to the existing security monitoring system to realize the intelligent management of the security monitoring system,and ensure the grain depot vehicles to operate in the important checkpoints of the grain depot in order.But the current environment of the grain depot is relatively complex(more dust,more checkpoints),there is a lot of noise in the collected image of grain depot vehicles,so it is difficult to deal with it by using the traditional vehicle recognition method.Based on the above background,this thesis studied the license plate location,license plate recognition and vehicle color recognition of grain depot vehicles in combination with the characteristics of grain depot.The main contents of this thesis are as follows:(1)This paper studied the license plate location algorithm based on the Maximally Stable Extremal Regions(MSER).Firstly,according to the characteristics of the MSER,this location method extracts the spot areas in the image,and filters these spot areas with a series of thresholds,the text characters are directly separated from the image spots,and the license plate area in the image is finally determined..After the experimental analysis and comparison,There is a conclusion that this algorithm is more suitable for the complex environment such as grain depot.(2)A 14 layer Convolutional Neural Networks(CNN)model is designed,and the image data set of the license plate of grain depot wagon is established to train it.The input grain depot complete license plate is recognized directly,which avoids the complex processing operations of license plate correction and character segmentation in the traditional method,and solves the problem that the license plate characters are not easy to be segmented in the traditional method under the condition of grain depot noise.Compared with the convolution neural network of nine layers and twelve layers,the accuracy of license plate recognition by these two comparison networks is 82.30% and 91.70% respectively,and the accuracy of the 14 layers network designed in this paper is 95.90%,which verifies the validity of the 14 layers convolution neural network.(3)This thesis studied the algorithm of grain depot vehicle color recognition based on subdivision region of interest(ROI),In the preprocessing stage,the determination and processing methods of ROI are improved,the determined ROI is divided into multiple small regions,and then converts them into HSV color space for recognition,and final result is determined,so as to reduce the impact of the structure of the grain depot vehicle and the external environment on the color recognition of the grain depot vehicle,the accuracy of grain depot vehicle color recognition is improved.According to the characteristics of grain depot,the algorithm of grain depot vehicle identification proposed in this thesis can effectively reduce the impact of dust environment of grain depot on the identification,improve the accuracy of grain depot vehicle identification,and is of great significance to the intelligent construction of grain depot security system.