Intelligent Survey Method for Tiny Rice Pests and Their Natural Predators in Paddy Fields Using Augmented Reality (AR) Glasses

Rice crops are frequently threatened by pests such as rice planthoppers ( Nilaparvata lugens , Sogatella furcifera , and Laodelphax striatellus ) and leafhoppers ( Cicadellidae ), which cause significant yield losses. Accurate identification of both pest developmental stages and their natural predators is crucial for effective pest control and maintaining ecological balance. However, conventional field surveys are often subjective, inefficient, and lack traceability. To overcome these limitations, this study proposed RiceInsectID, a two-stage cascaded detection method designed to identify and count tiny rice pests and their natural predators from white flat plate images captured by head-worn AR glasses. The method recognizes 25 insect classes, including 17 instars of rice planthoppers, 2 instars of leafhoppers, 4 spider species ( Araneae ), as well as Miridae and rove beetles ( Staphylinidae Latreille). At the first coarse-grained detection stage, 16 visually similar classes are consolidated into 6 broader categories and detected using an enhanced YOLOv6 model. To improve small object detection and address class imbalance, the full-region overlapping sliding slices and target pasting (FOSTP) algorithm was applied, increasing the mean average precision at a 50% IoU threshold (mAP 50 ) by 35.46% over the baseline YOLOv6. Feature extraction and fusion were further improved by incorporating an efficient channel attention path aggregation feature pyramid network (ECA-PAFPN) and adaptive structure feature fusion (ASFF) modules, while the balanced classification mosaic (BCM) enhanced detection of minority classes. With test-time augmentation (TTA), mAP 50 improved by an additional 2.06%, reaching 84.71%. At the second fine-grained classification stage, each of the six broad classes from the first stage is further classified using individual ResNet50 models. Online data augmentation and transfer learning were employed to significantly enhance generalization. Compared with the baseline YOLOv6, the two-stage cascaded method improved recall by 4.06%, precision by 3.79%, and the F1-score by 3.92%. Overall, RiceInsectID achieved 82.85% recall, 80.62% precision, and an F1-score of 81.72%, demonstrating an efficient and practical solution for monitoring tiny rice pests and their natural predators in paddy fields. This study provides valuable insights for ecosystem monitoring and supporting sustainable pest management in rice agriculture.