A consistent orientation of ginger shoots when sowing ginger is more conducive to high yields and later harvesting. However, current ginger sowing mainly relies on manual methods, seriously hindering the ginger industry’s development. Existing ginger seeders still require manual assistance in placing ginger seeds to achieve consistent ginger shoot orientation. To address the problem that existing ginger seeders have difficulty in automating seeding and ensuring consistent ginger shoot orientation, this study applies object detection techniques in deep learning to the detection of ginger and proposes a ginger recognition network based on YOLOv4-LITE, which, first, uses MobileNetv2 as the backbone network of the model and, second, adds coordinate attention to MobileNetv2 and uses Do-Conv convolution to replace part of the traditional convolution. After completing the prediction of ginger and ginger shoots, this paper determines ginger shoot orientation by calculating the relative positions of the largest ginger shoot and the ginger. The mean average precision, Params, and giga Flops of the proposed YOLOv4-LITE in the test set reached 98.73%, 47.99 M, and 8.74, respectively. The experimental results show that YOLOv4-LITE achieved ginger seed detection and ginger shoot orientation calculation, and that it provides a technical guarantee for automated ginger seeding.

Consistent ginger shoot orientation helps to ensure consistent ginger emergence and meet shading requirements. YOLO v3 is used to recognize ginger images in response to the current ginger seeder’s difficulty in meeting the above agronomic problems. However, it is not suitable for direct application on edge computing devices due to its high computational cost. To make the network more compact and to address the problems of low detection accuracy and long inference time, this study proposes an improved YOLO v3 model, in which some redundant channels and network layers are pruned to achieve real-time determination of ginger shoots and seeds. The test results showed that the pruned model reduced its model size by 87.2% and improved the detection speed by 85%. Meanwhile, its mean average precision (mAP) reached 98.0% for ginger shoots and seeds, only 0.1% lower than the model before pruning. Moreover, after deploying the model to the Jetson Nano, the test results showed that its mAP was 97.94%, the recognition accuracy could reach 96.7%, and detection speed could reach 20 frames·s⁻¹. The results showed that the proposed method was feasible for real-time and accurate detection of ginger images, providing a solid foundation for automatic and accurate ginger seeding.

Green onion (Allium fistulosum L.) is mainly available as factory-produced seedlings. Although factory seedling production is highly automated, miss-seeding during the seeding process considerably affects subsequent transplanting and the final yield. To solve the problem of miss-seeding, the current main method is manual complementary seeding, which is labor-intensive and inefficient work. In this study, an automatic machine-vision-based complementary seeding device was proposed to reduce the miss-seeding rate and as a replacement of manual complementary seeding. The device performs several main functions, including the identification of miss-seeding holes, control of seed case movement, and the seed uptake and release from the seed suction nozzle array. A majority-mechanism-based miss-seeding tray hole rapid-detection method was proposed to enable the real-time identification of miss-seeding tray holes in the tray under high-speed moving conditions. The structural parameters of the vacuum-generated seed suction nozzle were optimized through numerical simulations and orthogonal experiments, and the seed suction nozzle array and seed case were produced using 3D-printing technology. Finally, the complementary seeding device was installed on the tray-type green onion seeding machine and the effectiveness of the complementary seeding was confirmed by experiments. The results revealed that the average values of the precision, recall, and F1 scores for identifying miss-seeding tray holes were 98.48%, 97.00%, and 97.73%, respectively. The results revealed that the rate of miss-seeding tray holes decreased from 5.37% to 0.89% after complementary seeding.

The prediction of remaining useful life (RUL) of lithium–ion batteries is key to the reliability assessment of batteries and affects safe application of batteries. This article introduces a CEEMDAN-RF-MHA-ED-LSTM method. Using CEEMDAN, the battery capacity data were decomposed to obtain intrinsic mode functions (IMFs), and the weight of each IMF was obtained via the random forest (RF) algorithm. The LSTM neural network was used, the encoder–decoder (ED) structure was introduced, the multi-head attention (MHA) mechanism was used to construct a network model, and the particle swarm optimization (PSO) algorithm was used to optimize the model parameters. Each IMF was input into the model, and the obtained forecast results were weighted and reconstructed to obtain the final forecast data. This method was validated on the battery dataset released by NASA. Compared with that of the single LSTM model, the mean absolute error of the proposed method decreases by 74%, 62%, 71%, and 55% on the No. 05, 06, 07, and 18th battery datasets, respectively. The root mean square error decreased by 72%, 59%, 70%, and 54%, and the mean absolute percent error decreased by 75%, 65%, 71%, and 58%, respectively. This method can accurately predict battery RUL.

Design and test of the plant-correcting reel for harvesting lodging garlic plants

Fruit size is a key trait in small-fruit breeding, yet its measurement remains labor intensive and prone to human error. To address this, we developed a non-destructive, automated size measurement system based on machine vision and LabVIEW, designed for small fruits such as cherry, blueberry, and walnut. The system integrates a modular architecture, including flat-field correction, calibration, and pattern matching sub-VIs, to ensure user-friendly operation. These sub-VIs also enable the system’s core data analysis, such as real-size conversion through calibration and noise reduction for data accuracy through flat-field correction. An optimized image processing pipeline (grayscale conversion, Canny edge detection, morphological operations) enables precise contour extraction, even for fruits with stems or irregular surfaces. The system supports multi-species adaptation through lightweight parameter adjustments, without hardware modification. Experiments involved 15 samples per species (cherry ‘Tieton’, blueberry ‘Northland’, walnut ‘Xiangling’). A gold-standard protocol was established using a pre-calibrated digital caliper operated by two experienced technicians, with the mean of six replicates per fruit defined as the true value. Results demonstrated low root mean square errors, with coefficients of determination (R²) exceeding 0.98. Paired t-tests confirmed no significant differences from the gold standard. The system achieved a measurement speed of 0.4 s per fruit, six times faster than manual methods, and complied with the precision requirements of GB/T 26906-2024 (Sweet Cherry). This system offers a cost-effective, high-throughput solution for fruit breeding and phenotyping, effectively overcoming the limitations of manual measurement.