Research On Vehicle Detection Method In Remote Sensing Image Based On Deep Convolution Neural Network

With the continuous development of artificial intelligence,various algorithms of deep learning are gradually applied to every field of daily life.Among them,image detection is the most widely used,and the recognition and classification of remote sensing satellite images have unique value and significance in many fields.This paper mainly studies the target detection of vehicles in remote sensing images,and its application can effectively identify the vehicle target in the remote sensing image Its application can effectively identify the vehicle target in the remote sensing image.It can be widely used in traffic management,route planning and traffic flow detection.Therefore,based on the YOLOv3 deep learning neural network,this paper improves the target detection which is suitable for the characteristics of remote sensing image,and makes the yolv3 neural network model more accurate in the detection of small targets in the image.The method model proposed in this paper can be divided into two parts:First,according to the large size of satellite remote sensing image and the small scope of the target to be detected in the overall image,an improved neural network model based on YOLOv3 is proposed,which makes the original grid processing of YOLOv3 with large divided area become suitable for the detection of small and medium targets in remote sensing image.It is suitable for small target detection in remote sensing image.Then K-means algorithm is used to calculate the grid size of Yolo network model which is suitable for vehicles.Then YOLOv3 is used to train the data set to get the network model which is used to detect vehicles in remote sensing image.Finally,by testing the model.the vehicle detection model which is suitable for remote sensing image is obtained.Experimental results show that this method can effectively detect vehicles in remote sensing image and has high accuracy.Secondly,for the problem that the vehicle target in the remote sensing image is covered by trees or houses,The improved scheme proposed in the first part has a very good detection effect on the vehicle target without obstructions in the practical application of the improved network model of YOLOv3,however,it is not ideal for vehicles or houses covering more than 50% of the area covered by objects.Therefore,based on the improved scheme proposed in the first part,this paper proposes a new method,which can effectively detect blocked vehicle targets.This paper proposes a vehicle detection method in remote sensing images based on image preprocessing,which can effectively detect blocked vehicle targets.In order to strengthen the vehicle target in the remote sensing image,the remote sensing images in the experimental training set and test set will be preprocessed,and the data set will be processed for dilation and erosion before entering the network model,and then the YOLOv3 network model is improved,and multi-scale and multi-level feature fusion is added to it,which further strengthens the accuracy requirements of the network model for small target detection and improves the accuracy of vehicle detection in remote sensing images.To sum up,the improved YOLOv3 network model is better than the original YOLOv3 model for the detection of small vehicle targets in high-resolution satellite remote sensing images.Through the experimental analysis,it can be proved that the effectiveness and detection accuracy of this method are further strengthened,which means that the improved YOLOv3 network model has certain research significance.