The yield and security of grain are seriously infringed on by crop diseases, which are the critical factor hindering the green and high-quality development of agriculture. The existing crop disease identification models make it difficult to focus on the disease spot area. Additionally, crops with similar disease characteristics are easily misidentified. To address the above problems, this paper proposed an accurate and efficient disease identification model, which not only incorporated local and global features of images for feature analysis, but also improved the separability between similar diseases. First, Transformer Encoder was introduced into the improved model as a convolution operation, so as to establish the dependency between long-distance features and extract the global features of the disease images. Then, Centerloss was introduced as a penalty term to optimize the common cross-entropy loss, so as to expand the inter-class difference of crop disease characteristics and narrow their intra-class gap. Finally, according to the characteristics of the datasets, a more appropriate evaluation index was used to carry out experiments on different datasets. The identification accuracy of 99.62% was obtained on Plant Village, and the balanced accuracy of 96.58% was obtained on Dataset1 with a complex background. It showed good generalization ability when facing disease images from different sources. The improved model also balanced the contradiction between identification accuracy and parameter quantity. Compared with pure CNN and Transformer models, the leaf disease identification model proposed in this paper not only focuses more on the disease regions of leaves, but also better distinguishes different diseases with similar characteristics.

S-ovalbumin content is an indicator of egg freshness and has an important impact on the quality of processed foods. The objective of this study is to develop simplified models for monitoring the S-ovalbumin content of eggs during storage using hyperspectral imaging (HSI) and multivariate analysis. The hyperspectral images of egg samples at different storage periods were collected in the wavelength range of 401–1002 nm, and the reference S-ovalbumin content was determined by spectrophotometry. The standard normal variate (SNV) was employed to preprocess the raw spectral data. To simplify the calibration models, competitive adaptive reweighted sampling (CARS) was applied to select feature wavelengths from the whole spectral range. Based on the full and feature wavelengths, partial least squares regression (PLSR) and least squares support vector machine (LSSVM) models were developed, in which the simplified LSSVM model yielded the best performance with a coefficient of determination for prediction (R²_P) of 0.918 and a root mean square error for prediction (RMSEP) of 7.215%. By transferring the quantitative model to the pixels of hyperspectral images, the visualizing distribution maps were generated, providing an intuitive and comprehensive evaluation for the S-ovalbumin content of eggs, which helps to understand the conversion of ovalbumin into S-ovalbumin during storage. The results provided the possibility of implementing a multispectral imaging technique for online monitoring the S-ovalbumin content of eggs.

Soluble solids content is the most important attribute related to the quality and price of apples. The objective of this study was to detect the soluble solids content (SSC) in ‘Fuji’ apples using hyperspectral imaging combined with a deep learning algorithm. The hyperspectral images of 570 apple samples were obtained and the whole region of apple sample hyperspectral data was collected and preprocessed. In addition, a method involving multi-attention convolutional neural network (MA-CNN) is proposed, which extracts spectral and spatial features from hyperspectral images by embedding channel attention (CA) and spatial attention (SA) modules in a convolutional neural network. The CA and SA modules help the network adaptively focus on important spectral–spatial features while reducing the interference of redundant information. Additionally, the Bayesian optimization algorithm (BOA) is used for model hyperparameter optimization. A comprehensive evaluation is conducted by comparing the proposed model with CA-CNN models, SA-CNN, and the current mainstream models. Furthermore, the best prediction performances for detecting SSC in apple samples were obtained from the MA-CNN model, with an ${\mathrm{R}}_{\mathrm{p}}^2$ value of 0.9602 and an RMSEP value of 0.0612 °Brix. The results of this study indicated that the MA-CNN algorithm combined with hyperspectral imaging technology can be used as an effective method for rapid detection of apple quality parameters.

Complex harvesting environments and varying crop conditions often lead to threshing cylinder blockage and increased entrainment loss in maize grain harvesters. To address these issues, an electric-driven automatic control system for maize threshing concave clearance based on real-time entrainment loss monitoring was developed. The system automatically adjusts concave clearance parameters at different harvesting speeds to maintain grain entrainment loss within an optimal range. First, an adjustable concave structure based on a crank-link mechanism was designed, with a threshing clearance adjustment range of 15–47 mm and motor rotation angle of 0–48°. Subsequently, an EDEM simulation model of the mixed material discharge inside the threshing cylinder was established to determine the optimal installation position of the entrainment loss monitoring sensor based on piezoelectric ceramic-sensitive elements. The sensor was positioned at the left tail end of the concave sieve, with a minimum distance of 58 mm between the sensitive plate centerline and threshing concave sieve and an installation angle of 65° relative to the horizontal plane. A maize threshing clearance control method based on fuzzy neural network PID control algorithm was proposed, and Simulink simulation optimization verified its superior performance with fast response speed. After system integration, field trials were conducted at low, medium, and high operating speeds with preset ideal entrainment loss intervals. The results showed that control was unnecessary at low speed, the control system-maintained entrainment loss within set range at medium speed, and maximum threshing clearance was needed at high speed. Finally, comparative trials of threshing performance with and without the control system were conducted at medium harvesting speed. Results showed that the entrainment loss rate decreased by 43.75% with the control system activated, significantly reducing maize threshing entrainment losses. This study overcame the barrier of maize threshing parameter adjustment being heavily reliant on manual experience and provided theoretical support for the intelligent grain harvesting equipment.

The yield assessment process during maize harvesting is a necessary means to ensure farmers’ economic benefits and stable agricultural production. Predicting the mass of maize kernels is an important condition for yield detection. This study proposes a maize kernel mass prediction model based on machine vision and machine learning algorithms to determine whether the kernels are broken. By extracting the geometric features of maize kernels, a phenotypic feature dataset of maize kernels was constructed. Subsequently, popular machine learning algorithms were used to establish regression models for maize kernel mass, achieving quantitative prediction of maize kernel mass. The results indicate that the PLSR (Partial Least Squares Regression) and RF (Random Forest) algorithms are suitable for constructing mass prediction models for broken and unbroken kernels, respectively. The models established by the two algorithms achieved R-values of 0.941 and 0.925, respectively. Field trial results show that there is a strong linear relationship between the predicted maize kernel mass using the constructed model and the actual kernel mass. Therefore, this method can serve as an accurate, objective, and efficient detection method for maize yield.

Ensuring food safety requires the detection of trace-level contaminants such as pesticides, mycotoxins, and heavy metals. Analytical approaches for these analytes should feature high sensitivity, good selectivity, and compatibility with aqueous matrices; surface-enhanced Raman spectroscopy (SERS) satisfies these requirements. Addressing the absence of a unified comparative analytical framework, this critical review surveys recent SERS-enabled sensing strategies for food contaminants. Detection strategies differ substantially across the three contaminant classes: pesticides can be directly detected at ppb levels through substrate engineering and deep learning; mycotoxins rely on affinity-recognition elements to reach pg-mL-level sensitivity; and Raman-inactive heavy metals demand indirect readout via functional probes. Crucially, despite these divergent analytical routes, the field confronts three shared bottlenecks—spectral irreproducibility, severe matrix interference, and the lack of standardized protocols, all of which hinder regulatory adoption. Compared with near-infrared spectroscopy (NIR) and hyperspectral imaging (HSI), SERS delivers outstanding sensitivity for confirmatory trace-level analysis, while its limited throughput may be compensated by multispectral data fusion. Future advances should prioritize portable sensing hardware, explainable Artificial Intelligence (AI), and multiplexed detection to transfer laboratory-scale sensitivity toward practical field-deployable testing tools.