Optimization of Infrared Image Detection Technology Based on Deep Learning
Infrared target detection has the advantages of all-weather,long detectable distance,and not affected by atmospheric and lighting conditions.It is particularly suitable for unmanned vehicles in low light conditions at night and can achieve good detection results,with broad application prospects.However,infrared images suffer from low resolution,blurred edges,and poor contrast,leading to a decrease in detection accuracy.Therefore,based on the Flir dataset,the first step is to use linear spatial domain filtering technology to enhance infrared images,improve image edge clarity,and train images using the YOLOv5 algorithm deployed with BiFPN feature network.The results showed that the accuracy of the enhanced infrared image training model increased by 12.1%,the recall rate increased by 6.7%,the average accuracy increased by 20.7%at 0.5,and the average accuracy increased by 4.9%at 0.5:0.95,proving that this study can effectively improve the detection accuracy of infrared images.
NETL
NSTL
万方数据
