<p>Accurate fruit detection is vital significance for estimating the output of orchards. In this paper, an improved deep convolutional neural network detection model, SYL-YOLOv8n, suitable for small object detection from the perspective of unmanned aerial vehicles based on YOLOv8n is proposed. Firstly, this study designed a PE-C2f module, which founded on the partial convolution (PConv) and efficient multi scale attention mechanism (EMA). This design facilitates model lightweighting while improving the capacity to extract and integrate peach feature information in complex environments. Secondly, the multiple path distance intersection over union (MPDIoU) loss function is used as the new bounding box loss function to accelerate the positional fitting between the ground-truth and predicted boxes. Thirdly, the bidirectional feature pyramid network (BiFPN) is introduced to improve the information transmission capacity among features of different scales through weighted bidirectional feature fusion. Finally, a P2 small target detection layer is added in the neck section, thereby better capturing the detailed information of small objects. To verify the validity of the algorithm, experiments were carried out on a self-constructed multi-source peach dataset. The Precision (P), Recall (R), mean average precision (mAP), and F1-Score (F1) of the SYL-YOLOv8n reached 92.7%, 85.0%, 94.3%, and 88.7%, respectively, which were 1.4%, 1.9%, 0.7%, and 1.7% higher than the original model. Finally, based on the number of canopy fruits identified by SYL-YOLOv8n, a “canopy fruits - LAI - whole tree fruits” yield estimation method incorporating leaf area index (LAI) coefficients was proposed. The results showed that the R² of the yield estimation method reached 0.88, with a RMSE of 11.16. The accuracy of this estimation method also meets the actual application requirements.</p>