To address challenges in detecting small targets from aerial imagery and deploying models on edge devices for real-time detection, we propose the lightweight improvement, F5332SFLFST_YOLOv5s, based on YOLOv5s. The algorithm leverages the latest FasterNet network as an image feature extractor to enhance detection efficiency by reducing computational complexity. The native "1-2-8-2" structure is replaced with the explored "5-3-3-2" Backbone structure, incorporating more basic blocks early to extract shallow feature information. Streamlining the parameter passing of the FasterNet network further enhances the Backbone network's performance. By embracing the lightweight concept of FasterNet, a FasterC3 module is introduced at the end of the Backbone network to acquire more substantial high-level semantic information.additionally, combining the large selective kernel network, LSKNet, with the bottleneck structure simulates various ranging contexts in remote sensing scenarios by dynamically adjusting the spatial receptive field, thereby improving detection accuracy and efficiency within limited computational and memory resources. The Focal-αEIoU1.5 loss function is proposed, replacing the original CIoU loss function, adaptively improving IoU object loss and bbox regression accuracy through Power transform. Soft-NMS(Non-maximum suppression) is introduced as a substitute for traditional NMS, addressing information loss issues caused by the direct discarding of overlapping frames. Lastly, a three-level feature fusion structure, TriNeck, fuses neighboring shallow feature layers in the Backbone, obtaining more accurate shallow location features for precise target localization.experimental results on the VisDrone 2019 dataset show that compared to the YOLOv5s benchmark, our optimized lightweight model improves inference speed by 3%, with a 37.3% reduction in model parameters and a 37.5% reduction in computational load. Additionally, [email protected] and [email protected]:0.95 increase by 13.5% and 43.0%, respectively, indicating significant potential for real-time detection deployment on edge devices.