Multi-view infrared small target recognition based on YOLOv5

In order to improve the recognition accuracy of small infrared targets in complex environment, a SyluInfrared dataset is built by taking multi-view infrared images with the Frock TiX600 infrared thermal imager. Seven image enhancement algorithms including image brightening, contrast adjustment, image blurring, image flipping and image scaling are designed to improve the image quality of small infrared targets. The YOLOv5 deep learning model is used to identify small infrared targets with multiple views to improve the recognition accuracy of small infrared targets. The experimental results show that the accuracy of multi-view infrared small target recognition model for YOLOv5 deep network is 97.7%, which is 6.6% higher than that of single-view infrared small target recognition model based on YOLOv5. The multi-view target recognition strategy significantly improves the infrared small target recognition performance.