Improved YOLOv7-Tiny lightweight anti-UAV target detection algorithm

The use of Unmanned Aerial vehicles (UAVs) in both military and civilian contexts raises a series of safety issues. Traditional anti-UAV detection methods may cause false and missed detections when faced with complex backgrounds and large changes in target scale. To solve the above problems, a lightweight anti-UAV detection algorithm based on yolov7-tiny was proposed. Firstly, the YOLOv7-Tiny backbone network is substituted with an enhanced ShuffleNetV2 network, augmenting the network's feature extraction capacity to acquire richer semantic information while maintaining a lightweight model. Secondly, we propose a cross-scale fusion feature strategy inspired by the feature pyramid concept to enhance the model's detection capabilities across various target scales. Thirdly, the ELAN-DS module is introduced to reduce computations and parameters while maintaining detection effectiveness. Finally, the loss function is replaced with Focal-EIoU to improve detection accuracy for small targets. The improved algorithm is trained and evaluated on the UAV dataset. The [email protected] increases by 1.1%, the Parameters is reduced by 21.67%, the Floating Point Operations (FLOPs) is reduced by 33.33%, the model size is 10.4MB, and the frame rate reaches 48 frames per second, all in comparison to the YOLOv7-Tiny algorithm. The improved algorithm possesses a smaller model size and enhanced detection capability. Methods are provided for deployment on anti-UAV mobile devices or embedded systems.