A Lightweight Detection Model with Content-Aware Feature Reassembly for Passion Fruit Visual Maturity-Class Detection in Complex Orchard Environments

This study addresses the challenges of passion fruit visual maturity-class detection in complex natural orchard environments, particularly the missed detections of visually challenging fruit instances caused by background interference, occlusion, and illumination variations. To address this issue, an improved detection model, WEC-YOLO11, is proposed based on YOLO11n. The model incorporates the Efficient Channel Attention (ECA) mechanism to enhance discriminative feature representation while suppressing background interference, introduces the Content-Aware ReAssembly of Features (CARAFE) operator for content-aware upsampling, and adopts the WIoUv3 loss function with a dynamic non-monotonic focusing mechanism for bounding box regression. Across three independent runs, WEC-YOLO11 achieved an average mean average precision (mAP)@0.5 of 86.2% and an average recall of 81.1%, with 2.69 M parameters and 6.5 giga floating-point operations per second (GFLOPs). Compared to YOLO11n, the recall of ripe passion fruit (rpf) increased from 76.5% to 79.9% (+3.4 percentage points), while rpf precision showed a slight decrease from 77.6% to 77.1%, indicating a recall-oriented precision–recall trade-off. The overall recall increased from 79.9% to 81.1% (+1.2 percentage points). Under the tested batch-1 RTX 4090 setting, the measured throughput was 91.6 frames per second (FPS). These results indicate a favorable accuracy–complexity trade-off for passion fruit visual maturity-class detection within the evaluated public orchard dataset.