Pre-harvest protein and moisture content in wheat jointly determine market value and optimal harvest timing. However, existing detection methods are limited to expensive laboratory spectrometers or handheld devices requiring threshed grains, rendering practical in-field quality assessment inaccessible to smallholder farmers. Achieving low-cost, in-situ measurement requires overcoming three critical challenges: strong water absorption in fresh spikes masking weaker protein signals, morphological variations across cultivars causing spectral baseline shifts, and scattering-related distortions introduced by variable probe contact during handheld operation. To address these issues, we present WheatScout, a co-designed system integrating optimized optics with task-aware learning. The hardware features a diffuse optical path to minimize geometric bias, while the processing pipeline employs a moisture-first framework to explicitly decouple water interference prior to protein estimation. Furthermore, we integrate domain-invariant representation learning to enhance cross-cultivar generalization and a trainable scattering correction module to compensate for residual field-trial errors. During a 15-day harvest window, we systematically collected 600 wheat ear samples, covering 10 different varieties and five functional categories, thus constructing a dataset with broad spatiotemporal representativeness and varietal diversity. On this dataset, WheatScout achieved an R2 > 97% for moisture and an R2 > 92% for protein, which are comparable to lab-grade spectrometers at two orders of magnitude lower cost.

Flour adulteration poses significant health risks and economic losses for consumers, yet current detection methods are often impeded by high costs, limited sensitivity, and impractical laboratory requirements, leaving end users vulnerable. The difficulty is exacerbated by the need for extremely low detection limits, often at parts-per-million (ppm) levels, to identify adulterants. To address these challenges, we introduce FlourSpec, a low-cost and user-friendly system designed for on-site detection of flour adulteration at the ppm level. FlourSpec employs a spectral reconstruction algorithm to extract pertinent information from coarse-grained spectral data collected using a low-cost spectrometer. Recognizing that existing spectral reconstruction algorithms often struggle with significant errors that obscure trace adulterant information, we develop a novel end-to-end architecture that balances spectral fidelity and classification performance while minimizing the propagation of reconstruction errors. Additionally, we incorporate a hybrid attention mechanism to capture harmonic correlations within the spectra, effectively suppressing cross-band reconstruction errors. A supervised contrastive learning module is also incorporated to enhance the discriminability of trace features. Experimental evaluations demonstrate that FlourSpec achieves 97.44% accuracy in detecting multiple adulteration types and 93.12% accuracy in identifying various BPO concentrations. Notably, it is only 0.63% lower than expensive solutions and 18.74% higher than baseline systems at the same price, highlighting its effectiveness for trace-level detection.

Meat adulteration is a significant problem that can pose health risks economic losses to consumers. Current detection methods are hindered by high costs, limited capabilities, or time-consuming sample preparation, making them only accessible in laboratory tests and can not protect the safety of end-users. This paper introduces MeatSpec, a low-cost and user-friendly system for detecting meat adulteration using spectral imaging, to move the adulteration inspection out of laboratories. MeatSpec employs a multispectral camera to reduce costs while quickly capturing spectral images, but this leads to a decrease in spectral resolution and coverage. To solve this challenge, the system uses spectral reconstruction technology and innovative designs tailored for meat adulteration detection. This includes involving adulteration-related prior information during the reconstruction training phase and incorporating contrastive learning to enlarge the distances among reconstructed samples belonging to various adulteration types. Additionally, we devise distinct feature extractors for different bands based on characteristics of the reconstructed spectra and employ knowledge distillation to mitigate error in full-band reconstructed spectra while capturing features related to adulteration. Further, we extend our system to MeatSpec-G to improve its generalizability to varied adulteration conditions and unknown adulterants. To achieve this, we first propose a feature alignment-based training scheme to reduce the feature gap among samples of diverse concentrations and admixture patterns. Then, we propose a cascaded open-set recognition framework that decouples uncertainty quantification and anomaly feature discrimination, to address the limitations of softmax confidence in detecting distribution shifts and reconstruction artifacts. Experimental evaluations on 347 paired spectral images demonstrate that our system achieves a 91.06% accuracy in detecting multiple adulteration types, merely 7.78% inferior to the expensive professional solution, yet 21.58% superior to the baseline at the same price point. Moreover, our system can generalize to achieve an 88.89% detection accuracy in unknown adulteration conditions with a 27.78% improvement, and an 83.33% detection accuracy for unknown adulterants.

Meat adulteration is a significant problem that can pose health risks economic losses to consumers. Current detection methods are hindered by high costs, limited capabilities, or time-consuming sample preparation, making them only accessible in laboratory tests and can not protect the safety of end-users. This paper introduces MeatSpec, a low-cost and user-friendly system for detecting meat adulteration using spectral imaging, to move the adulteration inspection out of laboratories. MeatSpec employs a multispectral camera to reduce costs while quickly capturing spectral images, but this leads to a decrease in spectral resolution and coverage. To solve this challenge, the system uses spectral reconstruction technology and innovative designs tailored for meat adulteration detection. This includes involving adulteration-related prior information during the reconstruction training phase and incorporating contrastive learning to enlarge the distances among reconstructed samples belonging to various adulteration types. Additionally, we devise distinct feature extractors for different bands based on characteristics of the reconstructed spectra and employ knowledge distillation to mitigate error in full-band reconstructed spectra while capturing features related to adulteration. Experimental evaluations on 347 paired spectral images demonstrate that our system achieves a 91.06% accuracy in detecting multiple adulteration types, merely 7.78% inferior to the expensive professional solution, yet 21.58% superior to the baseline at the same price point.