Abstract: Yunnan Province is one of the major macadamia-producing regions in China, where large-scale cultivation under high-altitude ecological conditions has led to frequent off-season flowering and asynchronous flowering within the same tree. These phenomena reduce flowering synchrony, which is detrimental to pollination, fertilisation, fruit set, fruit retention, and uniform harvesting, thereby affecting yield stability and fruit quality. In addition, current flowering regulation practices still rely heavily on manual operations, which are labour-intensive, inefficient, costly, and highly dependent on subjective experience. As a result, traditional regulation approaches can no longer meet the urgent demand for precision-oriented and automated orchard management. To address these challenges and promote intelligent orchard management, this study proposes an intelligent identification method for macadamia inflorescences in natural environments based on an improved YOLO11n object detection model. The proposed model is specifically optimised to cope with several challenges encountered in real orchard environments, including complex background interference, multi-scale morphological variation of inflorescences, partial occlusion, and the constraints of real-time edge deployment. First, a Partial Multi-Scale Feature Aggregation (PMSFA) module is introduced. By integrating partial convolution with multi-scale feature extraction, this module efficiently captures and aggregates multi-source heterogeneous features from different network layers, thereby significantly enhancing the detection of small and easily occluded inflorescences in cluttered natural backgrounds. Second, to balance detection accuracy and computational efficiency, Group-Shuffle Convolution (GSConv) and the Variety of View Group Shuffle Cross-Stage Partial Network (VoV-GSCSP) are incorporated to construct a Slim-neck architecture. This design reduces computational complexity and parameter size while maintaining high detection accuracy, thus improving suitability for edge deployment. Third, in terms of loss function optimization, the original Complete Intersection over Union (CIoU) loss in YOLO11n is replaced with Focaler-PIoU v2. This modification increases the gradient contribution of high-quality samples during training and improves the stability of bounding-box regression, thereby enhancing localization accuracy under challenging conditions such as occlusion and scale variation. Extensive comparative experiments conducted on a dedicated dataset of Yunnan macadamia inflorescences demonstrate that the improved YOLO11n model consistently outperforms several mainstream state-of-the-art object detection methods across all evaluation metrics. Compared with Faster R-CNN, SSD, RT-DETR, YOLOv5s, YOLOv8n, YOLOv10n, and YOLOv12n, the proposed model achieves substantial performance gains. Specifically, Precision increases by 20.1, 5.4, 1.7, 3.5, 3.6, 2.0, and 0.8 percentage points, respectively, while Recall increases by 1.4, 16.2, 8.1, 5.0, 6.4, 9.1, and 6.7 percentage points, respectively. For mean average precision at an IoU threshold of 0.5 (mAP0.5), the proposed model achieves improvements of 6.4, 15.9, 5.7, 4.3, 4.1, 5.0, and 3.3 percentage points, respectively. Under the more stringent mAP0.5–0.95 metric, the corresponding improvements reach 20.1, 33.5, 11.9, 13.5, 10.6, 11.0, and 9.7 percentage points, respectively. These results indicate that the proposed method achieves superior precision, recall, and overall detection accuracy, particularly under complex natural conditions. To further verify the practical feasibility of the proposed method, the improved model is deployed on the Orange Pi AI Pro edge-computing platform. By utilising Neural Processing Unit (NPU)-accelerated inference together with multi-threaded scheduling optimization, the deployed system achieves a real-time detection speed of 72.8 frames per second (FPS), fully satisfying the requirements for accurate and real-time recognition in natural orchard environments. Overall, the results demonstrate that the proposed model achieves a favourable balance among detection accuracy, computational efficiency, and deployment feasibility, and provides strong technical support for the development of intelligent and precise flowering-period regulation equipment for macadamia production in Yunnan Province.