Research On Weak And Small Target Detection Technology From The Perspective Of UAV Based On Attention Mechanism

With the performance improvement of embedded devices and edge computing technology,it is more and more popular to configure deep learning programs on edge computing devices to complete visual tasks,which makes it possible for UAV to conduct high-altitude reconnaissance.Therefore,effective target detection has become the key.However,the current mainstream deep learning model can not well complete the target detection task in the UAV high-altitude reconnaissance scene,which is mainly caused by three reasons.Firstly,most of the current models are to solve the detection problem of conventional targets.However,the target in the UAV perspective is usually less than 30x30 pixels;Secondly,UAV is easy to encounter the interference of fog and haze in the process of high-altitude reconnaissance,and the existing target detection model can not detect effectively;Finally,the computing power of the embedded equipment carried by UAV is limited,and the computing power of the existing target detection model is too high to be deployed to the embedded equipment to realize the real-time detection task.This paper designs a lightweight network model based on attention mechanism and secondary feature fusion to solve the difficulties.Firstly,a lightweight backbone network is designed to reduces the amount of calculation of the model,making realtime detection possible;Secondly,the features extracted from the backbone network are fused from top to bottom and from bottom to top,which improves the detection effect of weak and small targets;Finally,the hybrid attention mechanism is added in the process of secondary feature fusion to improve the target detection effect in haze scene。The experimental results show that the accuracy and recall of weak and small target detection are improved by 4.9% and 9.4%;In the haze scene,the accuracy rate was improved by 2.1% and the recall rate was improved by 5.6%.To sum up,in order to complete the UAV high-altitude reconnaissance mission,a lightweight network model based on attention mechanism and secondary feature fusion module is designed to effectively solve the above problems.