Development and construction of infrared high-altitude small target detection and recognition analysis system based on YOLOv7

Based on the model development and construction of the yolo series, the target detection system in the infrared scene has been practiced several times in my previous articles. If you are interested, you can read it by yourself:

"Infrared Ocean Target Detection Practice, Recognizing Infrared Ocean Targets Based on Target Detection Model"

"Development and Construction of UAV Aerial Photography Vehicle Instance Segmentation Detection and Analysis System Based on YOLO in Infrared Scenes"

"Development and Construction of Ocean Target Detection and Recognition Analysis System Based on YOLO Integrated Model in Infrared Scenes"

"Development and construction of a high-altitude infrared target detection and recognition analysis system based on the perspective of a drone based on three lightweight YOLOv5n/s/m models, and compare the performance of the test analysis"

The commonality of these projects is that the targets in the infrared scene are relatively visible, and the size is not small. Although only one type of infrared target in this paper is an aircraft, the target visibility is very low and the scale is small, which is even more difficult to detect and identify. First look at the example effect:

Next look at the dataset:

 The YOLO format data annotation file is as follows:

 Here is the development and construction model based on yolov7. If you don’t know how to develop and build your own personalized application based on yolov7, you can refer to my ultra-detailed actual combat column, as follows:

"YOLOv7 builds a model from scratch based on its own data set, complete training, inference calculation ultra-detailed tutorial"

I won't go into details here.

The default iterative calculation of 100 epochs, simply look at the result indicators:

 Indeed, I did not expect that the detection effect of small targets in infrared scenes can be so good.