Real-time monitoring of moisture content during pea drying is required for the high quality of finished products and extending shelf life. However, conventional detection, such as oven drying, cannot fully meet the online industrial monitoring in modern agriculture, due to the destructive, time-consuming, and severe lag. Compared with conventional single-point near-infrared spectroscopy, hyperspectral imaging can be expected to reduce background noise under irregular shrinkage during pea drying. In this study, high-precision, non-destructive, and lightweight detection was proposed for pea moisture content using visible and near-infrared (Vis-NIR) hyperspectral imaging. Furthermore, 360 pea samples (variety "Zhongwan No. 6") were prepared and subjected to hot air drying treatment at 60 °C. Sample data was collected at nine time intervals to cover the full range from a fresh high-moisture to a dried low-moisture state. A dataset was then constructed with a representative moisture gradient. Hyperspectral images of all samples were acquired in the spectral range of 372.66 to 1 039.65 nm. A threshold segmentation (reflectance threshold of 0.15 to 1.0 at 857.53 nm) was applied to extract the region of interest (ROI), thereby avoiding human subjective errors. Raw spectral data were then preprocessed after image acquisition using standard normal variate (SNV) to eliminate optical path differences. The dataset was partitioned into a calibration set and a prediction set at a ratio of 3:1 using the Kennard-Stone algorithm. Furthermore, four algorithms of feature wavelength screening were compared, including least absolute shrinkage and selection operator (LASSO), bootstrapping soft shrinkage (BOSS), particle swarm optimization (PSO), and competitive adaptive reweighted sampling (CARS). Prediction models were constructed to select these features using partial least squares regression (PLSR), least squares support vector machine (LSSVM), categorical boosting (CatBoost), and lightweight one-dimensional convolutional neural networks (1D-CNN). Spectral response analysis revealed that the drying process induced a significant downward baseline drift over the entire band. According to the Kubelka-Munk theory, the physical shrinkage and surface roughness of peas were attributed to dramatically enhancing the diffuse scattering. Experimental results indicated that the CARS algorithm exhibited the optimal performance of feature dimensionality reduction. A subset of ten characteristic wavelengths accounted for only 5.7% of the full spectrum. An ablation study on feature wavelengths revealed that the CARS retained the high-correlation water absorption bands (e.g., around 970 nm) and low-correlation baseline reference points (e.g., 504.14 and 1 035.63 nm). Once the low-correlation baseline reference points were artificially removed in the ablation study, the prediction coefficient of determination of the model dropped drastically to 0.715 5. These variables were effectively compensated for the physical baseline drift during drying shrinkage, according to an implicit mathematical difference. Among them, the CARS-LSSVM non-linear model achieved the best overall performance, thereby yielding an of 0.964 8 and a root mean square error of prediction (RMSEP) of 0.047 7. The high accuracy was superior to the PSO-LSSVM model, which retained 77 features. Furthermore, the CARS-LSSVM model demonstrated a remarkable performance in computational efficiency; Its running time was recorded at 0.042 s on the same hardware platform, which was hundreds of times higher than the deep learning model (1D-CNN). In conclusion, the strategy was used to effectively extract pure chemical information via decoupling the physical-chemical coupling effects during drying. The CARS algorithm was employed to mine a minimalist feature subset with the LSSVM model. A robust solution was achieved to detect pea moisture content. This finding can also provide solid data support and theoretical guidance for the low-cost, low-power, and handheld multi-spectral intelligent sensors.

Apple morphological parameters are often required for the high efficiency and less physical damage to apple surface during measurement. However, the low accuracy or high hardware cost have limited to the reconstruction technology using three-dimensional vision. This study aims to propose an automatic measurement of apple surface three-dimensional reconstruction and morphological parameters using monocular fringe projection. High-precision measurement was also realized to significantly reduce the system complexity and economic cost. An economical and reliable technical path was then provided for the automatic detection and grading of fruits after harvest. A combined coding strategy was combined with the phase shift and gray code, in order to improve the high accuracy and anti-interference. Phase-shift code was used to achieve the high-precision phase recovery, while Gray code was to solve the ambiguity problem in phase unwrapping. Sub-millimeter reconstruction accuracy was then achieved in a single projection. The experimental results of standard sphere reconstruction show that the absolute error was only 0.29 mm, and the relative error was as low as 0.37%, indicating its excellent single-view reconstruction. A high-precision rotating platform was introduced to obtain the fully three-dimensional information of the apple surface. The self-occlusion was also avoided in a single view. The multi-angle rotation of apple samples was adjusted for the high-precision point cloud data of some surfaces from multiple perspectives. Then, the iterative nearest point algorithm was used to accurately register and mosaic the point cloud data from different perspectives. The geometric shape of the occluded area was effectively restored to finally construct a complete and seamless three-dimensional digital model of the apple surface. The automatic extraction of morphological parameters was further developed for the reconstructed apple point cloud model. The key appearance quality indexes of apples were accurately calculated, including the deformity index, the maximum transverse diameter and volume, indicating the uniformity of fruit shape. A series of experiments were carried out to evaluate the performance of the system. The standard plane and sphere were reconstructed after verification. The results show that there was the remarkable reconstruction accuracy of the system, where the root mean square error of the point cloud fitting of standard sphere was less than 0.095 mm, and the plane reconstruction error was 0.387 mm. Furthermore, 40 samples of red Fuji apples were measured to verify the effectiveness of the system in practical application. The deformity index, the maximum diameter and volume were determined by linear regression, compared with the standard values by manual measurement. There was a high degree of consistency between the automatic and the manual measurement. The fitting determination coefficients R² were 0.98, 0.97 and 0.98, respectively, and the root mean square errors were 0.53%(deformity index), 0.98 mm (maximum diameter) and 6.95 cm³ (volume), respectively. Especially in volume measurement, the average relative error was only 1.09%, indicating the high measurement accuracy. The monocular fringe projection shared the significant advantages, compared with the commonly-used three-dimensional reconstruction, such as linear structured light scanning or binocular stereo vision. The optical path structure of the system was greatly simplified to maintain the sub-millimeter high accuracy, thus effectively controlling the overall cost. This finding can provide an innovative solution with the high precision, low cost and easy integration, in order to realize the rapid acquisition of three-dimensional morphological parameters in the nondestructive testing of fruit external quality. There are the application potential and popularization in modern agricultural equipment, especially in the field of automatic sorting.

This study aims to improve the accuracy of the detection of adulterated minced beef with chicken, duck, and pork. A series of tests were carried out to quantitatively detect the adulterants in beef using hyperspectral reflectance (R) and transmittance (T) spectral data fusion. Firstly, three types of adulterants (chicken, duck, and pork) were added to the beef with a mass percentage of 2% to 50% concentration interval of 2%, respectively. Then the reflectance spectral data was collected to combine with SVM (support vector machine, SVM), random forest (RF), and long short-term memory (LSTM). A classification model was established to classify the samples into chicken, duck, and pork adulterated and pure beef samples (4 groups). Then, a partial least squares model (PLSR) was built from the single reflectance and transmittance spectral data to quantify the concentrations of adulterants. The performance of the model was optimized by competitive adaptive reweighted sampling (CARS), irrelevant variable elimination (UVE), and primary and intermediate data fusion. The predictions were also conducted on adulterants such as chicken, duck, and pork, according to the intermediate-level data fusion. The results showed the greatest effectiveness values of improvement were 3.9%, 9.3%, and 4.5% over those in the optimal model with single spectral data, respectively. The models with the best prediction results for chicken and duck samples withadulterants were the de trending (DT) pre-processed UVE-PLSR, with corresponding coefficients of determination (r-squaredprediction, ) and root mean square error (RMSEP) of 0.984 5 and 1.8651、 0.986 0 and 1.7711, respectively, and the bestprediction model for adulterated pork samples was RAW-CARS-PLSR, with corresponding and RMSEP of 0.975 1 and 2.366 5. Hyperspectral imaging combined with data fusion can effectively detect the adulterants in beef with high accuracy and speed. The intermediate-level data fusion can maximize the performance of the improved model.

Terahertz imaging has been one of the most promising technologies for non-contact detection in recent years. Among them, the internal quality of sunflower seeds can be detected using terahertz imaging. Specifically, terahertz waves can penetrate the shells of sunflower seeds, and then interact with their internal tissue structures and components, thereby obtaining their information of quality. However, the imaging speed has severely impeded the effective application of terahertz imaging in large-scale and rapid detection scenarios. Traditional terahertz imaging can usually consume a substantial amount of time to collect and then process the image data. Thus, it is highly required to accurately and rapidly detect the quality of sunflower seeds, in order to fully meet the high-efficiency requirements in industrial production or actual detection in modern agriculture. In this study, the compressed sensing with the Attention-Enhanced Super-Resolution Generative Adversarial Network (A-ESRGAN) model was proposed in the field of terahertz imaging. Rapid and accurate detection was also achieved in the plumpness of sunflower seeds. Furthermore, the crucial step was to select the measurement matrix in the initial stage during compressed sensing. Firstly, the Compressive Sampling Matching Pursuit (CoSaMP) reconstruction was adopted to verify the performance of different measurement matrices. According to the best overall performance, the Gauss matrix was then selected as the final measurement matrix. Secondly, the Peak Signal-to-Noise Ratio (PSNR) and reconstruction time were also compared using five reconstruction algorithms, such as the Alternating Direction Method of Multipliers (ADMM) combined with Total Variation (TV) regularization (ADMM_TV) and Subspace Pursuit (SP). Evaluation indicators show that the ADMM_TV performed best, in terms of the PSNR, Mean Squared Error (NMSE), and Structural Similarity Index (SSIM). The Natural Image Quality Evaluator (NIQE) shared the lowest value when the measurement ratio exceeded 6.0%. Although there was no outstanding advantage in the reconstruction time, the overall performance of the ADMM_TV algorithm was better than that of the rest. Finally, the Attention-Enhanced Super-Resolution Generative Adversarial Network with Multiple Scales (A-ESRGAN-Multi) model was utilized to process the reconstructed images with the different sampling rates during compressed sensing. A better performance was then achieved, compared with the Real-ESRGAN and the Attention-Enhanced Super-Resolution Generative Adversarial Network with a Single Scale (A-ESRGAN-Single). The image quality and the edge contrast were enhanced for the subsequent image segmentation. Consequently, it was very feasible to combine the compressed sensing and the A-ESRGAN-Multi model, in order to detect the fullness of sunflower seeds. The average error of plumpness on the validation set was only 2.50%, and the maximum error of detection was 6.41%. In conclusion, the combination of compressed sensing and the A-ESRGAN-Multi model can effectively save 82.5% of the sampling time and storage space. Data volume was also reduced to alleviate the pressure on the storage devices and hardware costs during data transmission and processing. Thus, the finding can also provide new technical support to the quality detection of sunflower seeds.