Real-time segmentation and phenotypic analysis of rice seeds using YOLOv11-LA and RiceLCNN.

Introduction: The real-time, accurate detection and classification of rice seeds are crucial for improving agricultural productivity, ensuring grain quality, and promoting smart agriculture. Although significant progress has been made using deep learning, particularly convolutional neural networks (CNNs) and attention-based models, earlier methods such as threshold segmentation and single-grain classification faced challenges related to computational efficiency and latency, especially in high-density seed agglutination scenarios. This study addresses these limitations by proposing an integrated intelligent analysis model that combines object detection, real-time tracking, precise classification, and high-accuracy phenotypic measurement. Methods: The proposed model utilizes the lightweight YOLOv11-LA for real-time grain segmentation, which builds upon the YOLOv11 architecture. YOLOv11-LA incorporates several enhancements over YOLOv11, including separable convolutions, CBAM (Convolutional Block Attention Module) attention mechanisms, and module pruning strategies. These modifications not only improve detection accuracy but also significantly reduce the number of parameters by 63.2% and decrease computational complexity by 51.6%. For classification, the model employs a custom-designed, lightweight RiceLCNN classifier. Additionally, the DeepSORT algorithm is employed for real-time multi-object tracking, and sub-pixel edge detection along with dynamic scale calibration mechanisms are applied for precise phenotypic feature measurement. Results: Compared to YOLOv11, the YOLOv11-LA model increases the mAP@0.5:0.95 score by 1.9%, showcasing its superior detection performance while maintaining lower computational overhead. The RiceLCNN classifier achieved classification accuracies of 89.78% on private datasets and 96.32% on public benchmark datasets. The system demonstrated high accuracy in measuring phenotypic features such as seed size and roundness, with measurement errors kept within 0.1 millimeters. The DeepSORT algorithm effectively managed multi-object tracking, reducing duplicate identifications and frame loss in real-time. Discussion: Experimental validation confirmed that the YOLOv11-LA model outperforms the original YOLOv11 in terms of both detection speed and accuracy, while also maintaining low computational complexity. The integration of the YOLOv11-LA, RiceLCNN, and DeepSORT algorithms, combined with advanced measurement techniques, underscores the model's potential for industrial applications, particularly in enhancing smart agricultural practices.