Innovative Proximal Optical Sensing Techniques and Methodologies for Food and Agricultural Products

Optical sensing techniques in the electromagnetic region between visible (VIS) light and terahertz (THz) can present real-time, accurate, high-speed, and high-throughput inspection of food and agricultural products based on physical and chemical characteristics. Instruments and associated algorithms/theories are becoming more sophisticated in this area. Thus, advanced research methodologies are needed. Methodologies for food and agricultural product inspection based on optical sensing techniques in VIS, NIR, and THz regions were studied from the five aspects: routine calibration, joint calibration, label-free measurement, super-resolution, and internal imaging. The theories, applications, and future trends of these methodologies were discussed. A sparse coefficients wavelength selection and regression (SCWR) model was proposed as the innovative routine calibration method in the present study. SCWR can rapidly and simultaneously operate regression and select wavelengths on spectral datasets with multiple response variables without any random procedure and cross-validation in the model. For the joint calibration, an attention-based spectra encoding-spectra/property decoding model for VIS-NIR spectral analysis was proposed. This model aimed to solve the problem: soil VIS-NIR spectral data collected worldwide cannot be combined to provide consistent calibration models due to the differences in instruments and measurement methods, reducing the benefit of producing soil spectra libraries. The THz-time domain spectroscopy is suitable for label-free sensing. Besides, THz radiation has not previously been applied to wood quality traits assessment of tree ring cores. To investigate the effectiveness of selective tree improvement work, an accurate and label-free framework based on THz time-domain sensing for intrinsic wood quality traits assessment was specifically designed for tree ring core sample analysis. This study contains a global phase unwrapping strategy, effective medium theory (EMT)-based measurement, combined measurement, and resolution enhancement, all of which aim to solve and discuss several challenges in THz time-domain sensing. A local pixel graph neural network was built for THz time-domain imaging super-resolution because the low image acquisition speed of THz time-domain imaging systems limits their application in biological products analysis. The method could be applied to the analysis of any heterogeneous biological products as it only required a small number of sample images for training and particularly it focused on THz feature frequencies. Mapping the chemical distribution of seed kernels inside shells is a challenging task. In the current study, a nondestructive THz time-domain imaging system in transmission mode was used for imaging the energy and moisture distributions of sunflower seed kernels inside shells. For this task, a dual autoencoders (AE)-generative adversarial net (GAN) spectral dehulling semi-supervised model was developed. The model could automatically learn the kernel information from the latent representations of the spectra of the intact seeds through adversarial learning to achieve feature disentanglement. Overall, this study developed innovative research methodologies for optical sensing techniques in food and agricultural product research. Obviously, they have contributed to the field by advancing research progress in routine calibration, joint calibration, label-free measurement, super-resolution, and internal imaging.