To address the current issues of low intelligence and accuracy in seed-sorting devices, an intelligent seed sorter was developed in this study using machine-vision technology and the lightweight YOLOv5n. The machine consisted of a transmission system, feeding system, image acquisition system, and seed screening system. A lightweight YOLOv5n model, FS-YOLOv5n, was trained using 4756 images, incorporating FasterNet, Local Convolution (PConv), and a squeeze-and-excitation (SE) attention mechanism to improve feature extraction efficiency, detection accuracy, and reduce redundancy. Taking ‘Zhengdan 958’ corn seeds as the research object, a quality identification and seed sorting test was conducted on six test groups (each consisting of 1000 seeds) using the FS-YOLOv5n model. Following lightweight improvements, the machine showed an 81% reduction in parameters and floating-point operations compared to baseline models. The intelligent seed sorter achieved an average sorting rate of 90.76%, effectively satisfying the seed-sorting requirements.

Traditional deep learning methods employing 2D images can only classify healthy and unhealthy seedlings; consequently, this study proposes a method by which to further classify healthy seedlings into primary seedlings and secondary seedlings and finally to differentiate three classes of seedling through a 3D point cloud for the detection of useful eggplant seedling transplants. Initially, RGB images of three types of substrate-cultivated eggplant seedlings (primary, secondary, and unhealthy) were collected, and healthy and unhealthy seedlings were classified using ResNet50, VGG16, and MobilNetV2. Subsequently, a 3D point cloud was generated for the three seedling types, and a series of filtering processes (fast Euclidean clustering, point cloud filtering, and voxel filtering) were employed to remove noise. Parameters (number of leaves, plant height, and stem diameter) extracted from the point cloud were found to be highly correlated with the manually measured values. The box plot shows that the primary and secondary seedlings were clearly differentiated for the extracted parameters. The point clouds of the three seedling types were ultimately classified directly using the 3D classification models PointNet++, dynamic graph convolutional neural network (DGCNN), and PointConv, in addition to the point cloud complementary operation for plants with missing leaves. The PointConv model demonstrated the best performance, with an average accuracy, precision, and recall of 95.83, 95.83, and 95.88%, respectively, and a model loss of 0.01. This method employs spatial feature information to analyse different seedling categories more effectively than two-dimensional (2D) image classification and three-dimensional (3D) feature extraction methods. However, there is a paucity of studies applying 3D classification methods to predict useful eggplant seedling transplants. Consequently, this method has the potential to identify different eggplant seedling types with high accuracy. Furthermore, it enables the quality inspection of seedlings during agricultural production.

To address the issues of time-consuming, laborious, and inaccurate traditional detection methods of rice appearance quality, a complete set of rice appearance quality inspection schemes is proposed in this study based on the YOLOv8 object detection model. A method of detecting the whole kernel rate by using the rice image area was proposed that is based on the YOLOv8 detection model. This method is more efficient and faster than the traditional method of measuring the whole kernel rate by the weight of rice. The detection of yellow grains quality based on the color threshold and chalkiness degree of rice based on the gray threshold was realized. Finally, comparative tests were conducted to verify the effectiveness and feasibility of the detection method. Random rice samples were tested by manual testing methods following national standards. The results show that the whole kernel rate is 59.80%, the roughness rate is 78.10%, the mold rate is 4.32%, the chalkiness degree is 11.00%, and the grade is second-class indica rice. The same rice sample was tested using the method of this study. The results show that the whole kernel rate of 59.10%, the roughness rate of 77.30%, the mold rate of 4.20%, the chalkiness degree of 12.00%, and the grade is second-class indica rice. The results are basically consistent with the manual detection method of the national standard. The rice appearance quality detection method proposed in this study provides a reference for achieving efficient, non-destructive, accurate detection and preferential grading of rice quality.

Background: Accurate phenotyping of breast, drumstick, and wing yield is essential for genetic improvement in poultry breeding. However, these key economic traits can traditionally be measured only after slaughter, forcing breeding programs to rely on time-consuming, costly, and operator-biased sibling testing. This has become a major challenge for large-scale, high-precision, and non-destructive phenotypic evaluation in commercial poultry breeding. In this study, a non-destructive in vivo phenotyping framework based on digital radiography (DR) imaging combined with deep learning and machine learning methods is established. Results: A standardized DR image acquisition platform is constructed for in vivo phenotyping of live chickens. A lightweight multi-objective segmentation network, DRSegNet, is designed to achieve precise segmentation of breast, drumstick, and wing regions, with Dice coefficients higher than 97%. Morphological features were extracted from the segmentation masks and fed into machine learning models for weight regression. Five-fold cross-validation revealed strong predictive performance: the best-performing models achieved mean Pearson correlation coefficients of 0.8972, 0.7696, and 0.8049 for breast, drumstick, and wing weights, respectively. A phenotypic index (PI) and carcass balance index (CBI) are further constructed to enable comprehensive and balanced multi-trait selection. The deployed system supports a complete detection cycle of approximately 40 s per chicken, allowing high-throughput in vivo detection. Conclusions: This DR-based non-destructive framework enables reliable, real-time, and high-throughput in vivo prediction of key carcass traits in live chickens. It transforms traditional post-slaughter measurements into in vivo phenotypes obtainable directly from individual birds, may improve the accuracy and efficiency of breeding selection, and has the potential to shorten the breeding cycle. This study provides a practical and scalable technical solution for intelligent phenotyping in poultry, and may contribute to data-driven precision breeding systems.