Research On Target Detection Algorithm Of Remote Sensing Ship Image In Weak Light Environment Based On HB-YOLO

In recent years,with the continuous innovation of computer hardware communication technology,it provides technical support for the application of remote sensing ship target detection.At present,the ship image produced by high-resolution remote sensing satellites contains more feature information,but the traditional target detection method is powerless.Compared with traditional methods,the accuracy,quantity,and time of detection of deep learning are improved.At the same time,since remote sensing satellites receive ground radiation sensors in reality,the information collected is susceptible to interference from various factors between ground and air,such as the change of day and night light,rain and snow weather,etc.,which will bring difficulties to ship detection and identification.Therefore,in the face of image interference and remote sensing ship identification and detection tasks,this thesis carries out research work by combining remote sensing ship detection and identification with the improved YOLOV5 algorithm model under the interference of a weak light environment.Therefore,the work in this thesis is divided into two stages: first,the interference image is preprocessed to restore it,and then the improved deep learning algorithm is used to detect and identify remote sensing ships.The contents of this thesis are as follows.First,from the perspective of the remote sensing imaging principle,this thesis firstly carries out light compensation for remote sensing images in weak light environments,to realize remote sensing image restoration in a weak light environments.All optical remote sensing satellites are passive and imaging is heavily affected by light conditions.Based on this,aiming at the problem of poor imaging of visible remote sensing satellites in the low light environment,a light compensation network is constructed in this thesis,which is used for light compensation of remote sensing images in the low light environment and image restoration for images with different light brightness.The experiment verifies that the image enhancement network has a good recovery effect.Second,conduct remote sensing ship image target detection.After studying and analyzing the principle and characteristics of the YOLO series detection algorithm,the YOLOV5 model is improved specifically.There are two main angles: First,although the general convolution used by YOLOV5 can capture the interaction between adjacent pixels of the image,deep interactions may be ignored when processing complex images.Therefore,the recursive gate convolution module with high-order spatial interaction was introduced in this thesis to construct the HB-YOLO model,and the recall rate R was increased by 2.8% through the training model compared with that before the improvement.m AP＿0.5 increased by 1% to85.6%.Secondly,the attention mechanism and the large convolution kernel mechanism are introduced into the YOLOV5 model to build the RC-YOLO model,which provides a good perception of the global feature information.Under the combined action of the two,the recall rate R is increased by 2.4%.m AP＿0.5 increased from 1.7% to 86.3%.Finally,this thesis preprocesses the image in the weak light environment and then inputs it into the improved network model,and obtains good detection results.