The demand for safe and edible meat has led to the advancement of freeze-storage techniques, but falsely labeled thawed meat remains an issue. Many methods have been proposed for this purpose, but they all destroy the sample and can only be performed in the laboratory by skilled personnel. In this study, hyperspectral image data were used to construct a machine learning (ML) model to discriminate between freshly refrigerated, long-term refrigerated, and thawed beef meat samples. With four pre-processing methods, a total of five datasets were prepared to construct an ML model. The PLS-DA and SVM techniques were used to construct the models, and the performance was highest for the SVM model applying scatter correction and the RBF kernel function. These results suggest that it is possible to construct a prediction model to distinguish between fresh and non-fresh meat using the spectra obtained by purifying hyperspectral image data cubes, which can be a rapid and non-invasive method for routine analyses of the meat storage state.

This study developed non-destructive technology for predicting apple size to determine optimal harvest timing of field-grown apples. RGBD images were collected in field environments with fluctuating light conditions, and deep learning techniques were integrated to analyze morphometric parameters. After training various models, the EfficientDet D4 and Mask R-CNN ResNet101 models demonstrated the highest detection accuracy. Morphometric metrics were measured by linking boundary box information with 3D depth information to determine horizontal and vertical diameters. Without occlusion, mean absolute percentage error (MAPE) using boundary box-based methods was 6.201% and 5.164% for horizontal and vertical diameters, respectively, while mask-based methods achieved improved accuracy with MAPE of 5.667% and 4.921%. Volume and weight predictions showed MAPE of 7.183% and 6.571%, respectively. For partially occluded apples, amodal segmentation was applied to analyze morphometric parameters according to occlusion rates. While conventional models showed increasing MAPE with higher occlusion rates, the amodal segmentation-based model maintained consistent accuracy regardless of occlusion rate, demonstrating potential for automated harvest systems where fruits are frequently partially obscured by leaves and branches.

Growers need to assess fruit growth and harvest timing, and monitoring individual fruit volume informs both. However, this remains challenging on the plant because leaves occlude the fruit and a single viewpoint reveals only part of its surface. Three-dimensional (3D) shape completion can recover fruit geometry, and thereby volume, from partial point clouds. Most methods treat each observation independently and ignore the temporal continuity of repeated visits to the same fruit, so the estimated volume trajectories become noisy and unstable. We present SCoRe (Sequential CoRe), which extends a single-frame DeepSDF-based encoder with temporal latent fusion. The fusion averages the shape code across visits, favoring the latest observation as gaps grow, and keeps that code from moving backward along a growth axis found in the data, since a healthy fruit does not shrink before harvest. It introduces no learned parameters and runs at inference time. We evaluate SCoRe on 2,513 RGB-D frames of Korean melon (Cucumis melo L. var. makuwa, 210 fruits) imaged across two seasons and three greenhouses, using post-harvest photogrammetric meshes as ground truth and caliper measurements as an intermediate-frame reference. Against an identical single-frame encoder, SCoRe reduces the last-frame volume root mean squared error (RMSE) from 87.4 to 50.6 mL, a 42% reduction, and raises F1 at 5 mm from 66.1% to 83.0%. It surpasses established single-frame shape completion baselines, with the largest gains under heavy occlusion, and remains robust under irregular revisit intervals, as repeated crop monitoring requires. Code, trained weights, and data are publicly available.

Conventional crop water stress index (CWSI) methods face limitations in unmanned aerial vehicle (UAV) applications due to calibration requirements and intermittent observation constraints. This study developed an integrated approach combining thermal image calibration, deep learning-based canopy segmentation, and artificial neural networks (ANN) for crop water stress assessment in apple orchards. The framework comprised (i) ANN-based temperature calibration using a 300 mm × 300 mm blackbody with atmospheric variables as inputs, (ii) Mask R-CNN instance segmentation for canopy delineation, and (iii) an ANN-based water stress model predicting stomatal conductance directly from canopy temperature, air temperature, and relative humidity. Experiments were conducted under automated irrigation at four soil water tension set points (-20, -40, -60, and -80 kPa). UAV thermal imaging used a Zenmuse XT2 sensor at 30 m altitude, with concurrent ground measurements of stomatal conductance using an LI-600P porometer. The ANN-based temperature calibration achieved R² = 0.96 and RMSE = 0.68 °C, reducing the ground-to-aerial temperature discrepancy from 2.97 °C to 0.73 °C. Mask R-CNN achieved a mean average precision at IoU 0.5 (mAP@0.5) of 94.85% in canopy segmentation. The ANN-based water stress model achieved R² = 0.86 and RMSE = 0.107 mol m⁻² s⁻¹ in predicting stomatal conductance, compared with R² = 0.56 for the conventional empirical CWSI. The approach removes the need for repeated baseline establishment and provides tree-level assessment for precision irrigation management.