【Objective】Aiming at the characteristics of various target scales, complex background and dense small targets in aerial images of unmanned aerial vehicles (UAC), a small target detection algorithm LM-YOLO based on YOLOV5 is proposed.【Method】Firstly, the number of small target detection head was increased and K-DBSCAN clustering algorithm was used to optimize the anchor frame, so as to generate an anchor frame more suitable for small target detection and improve the detection accuracy of the algorithm. Then, a more efficient MobileNetV3-CBAM was designed as a feature extraction network to reduce the size of the network model. Finally, the large kernel selective attention mechanism LSK was introduced into the feature fusion network to increase the resolution of the model to similar targets.【Result】The experimental results on the public data set VisDrone2019 show that the average detection accuracy of LM-YOLO for all targets is improved by 7.6% and the model size is reduced by 45% compared with the benchmark model YOLOV5.【Conclusion】Experiments show that the proposed algorithm can reduce the model size while maintaining good detection accuracy.