Research And Design Of Ship Recognition System In Visible Image

This thesis introduces a ship identification and statistics system based on convolution neural network under visible light conditions,which is used to detect ships near the coast and ports.The convolution neural network has been widely used in the field of image recognition.We have studied the extensibility of convolution neural network in the field of ship recognition.Finally,this thesis realizes a real-time and accurate ship detection system.The scene of our experiment in this thesis is some port environments and nearcoastal areas.We count and identify ships within our monitoring range.At the same time,a ship identification statistical system based on convolution neural network is designed to provide visual interface for the port management department.Finally,we achieve intelligent management of port and coastal vessels,the main tasks of this thesis are:(1)For the ship image captured by the port,we propose an improved algorithm to remove fog by calculating the value of the dark channel to correct the color distortion,so as to improve the visual effect of the image.At the same time,the high brightness value of three color channels is reduced,and the mean value method is used to identify the image with better effect.On this basis,the bright area that does not conform to the dark channel is improved,making the processed image more clear and natural.The characteristics of the processed ship pictures are more obvious and have a good auxiliary effect on the identification and statistics of ships with subsequent convolution neural networks.This pretreatment method combined with convolution neural network improves the accuracy of ship detection.(2)Based on the morphological characteristics of the ship,we improve the structure of the convolutional neural network and make the structure more suitable for the scene of ship recognition.The adaptive network detection model for ship detection and identification based on the adaptive network of the existing convolution neural network has greatly improved the accuracy of the detection.This thesis has conducted experiments on the six most common types of ships,and has obtained relatively good results by collecting the video collected by the port camera and detecting and identifying the pictures of the ships.(3)We design and implement the ship identification statistical system based on convolution neural network under visible light conditions.We provide different services through scene recognition.We identify and count images of ships in both port video and drone aerial photos,and provide visual display.We can provide real-time detection statistics for port ships.