Maize seed variety classification has become essential in agriculture, driven by advancements in non-destructive sensing and machine learning techniques. This study introduced an efficient method for maize variety identification by combining hyperspectral imaging with a framework that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. Spectral data were acquired by hyperspectral imaging technology from five maize varieties and processed using Savitzky–Golay (SG) smoothing, along with standard normal variate (SNV) preprocessing. To enhance feature selection, the competitive adaptive reweighted sampling (CARS) algorithm was applied to reduce redundant information, identifying 100 key wavelengths from an initial set of 774. This method successfully minimized data dimensionality, reduced variable collinearity, and boosted the model’s stability and computational efficiency. A CNN-LSTM model, built on the selected wavelengths, achieved an accuracy of 95.27% in maize variety classification, outperforming traditional chemometric models like partial least squares discriminant analysis, support vector machines, and extreme learning machines. These results showed that the CNN-LSTM model excelled in extracting complex spectral features and offering strong generalization and classification capabilities. Therefore, the model proposed in this study served as an effective tool for maize variety identification.

Traditional biochemical analysis methods are not only resource-intensive and time-consuming, but are increasingly inadequate for meeting the demands of modern production and quality testing. In recent years, hyperspectral imaging (HSI) technology has been widely applied as a non-destructive detection method for fruit and vegetable quality assessment. This study, based on HSI technology, systematically investigates the distribution patterns of jujube quality parameters under various drying temperature conditions. It focuses on analyzing six key quality indicators: L*, a*, b*, soluble solid content (SSC), hardness, and moisture content. HSI was used to acquire reflectance (R), absorbance (A), and Kubelka–Munk (K-M) spectral data of jujubes at various drying temperatures, followed by several spectral preprocessing methods, including standard normal variate (SNV), baseline correction (baseline), and Savitzky–Golay first derivative (SG1st). Subsequently, a nonlinear support vector regression (SVR) model was used to perform regression modeling for the six quality parameters. The results demonstrate that the SG1st preprocessing method significantly enhanced the predictive capability of the model. For the predictions of L*, a*, b*, SSC, hardness, and moisture content, the best inversion models achieved coefficients of determination $R_p^2$ of 0.9972, 0.9970, 0.9857, and 0.9972, respectively. To further enhance modeling accuracy, deep learning models such as bidirectional long short-term memory (BiLSTM), bidirectional gated recurrent unit (BiGRU), and convolutional neural network–bidirectional gated recurrent unit (CNN-BiGRU) were introduced and compared comprehensively under the optimal spectral preprocessing conditions. The results demonstrate that deep learning models significantly improved modeling accuracy, with the CNN-BiGRU model performing particularly well. Compared to the SVR model, the $R_p^2$ values for L*, a*, b*, SSC, hardness, and moisture increased by 0.005, 0.007, 0.008, 0.011, 0.007, and 0.006, respectively; the $RPD$ values increased by 0.036, 0.04, 0.26, 0.462, 0.428, and 0.216. This study provides important insights into the further application of HSI technology in the quality monitoring and optimization of the jujube drying process.

Tea polyphenols, amino acids, soluble sugars, and other ingredients in fresh tea leaves are the key parameters of tea quality. In this research, a tea leaf ingredient estimation sensor was developed based on a multi-channel spectral sensor. The experiment showed that the device could effectively acquire 700–1000 nm spectral data of tea tree leaves and could display the ingredients of leaf samples in real time through the visual interactive interface. The spectral data of Fuding white tea tree leaves acquired by the detection device were used to build an ingredient content prediction model based on the ridge regression model and random forest algorithm. As a result, the prediction model based on the random forest algorithm with better prediction performance was loaded into the ingredient detection device. Verification experiment showed that the root mean square error (RMSE) and determination coefficient (R²) in the prediction were, respectively, as follows: moisture content (1.61 and 0.35), free amino acid content (0.16 and 0.79), tea polyphenol content (1.35 and 0.28), sugar content (0.14 and 0.33), nitrogen content (1.15 and 0.91), and chlorophyll content (0.02 and 0.97). As a result, the device can predict some parameters with high accuracy (nitrogen, chlorophyll, free amino acid) but some of them with lower accuracy (moisture, polyphenol, sugar) based on the R² values. The tea leaf ingredient estimation sensor could realize rapid non-destructive detection of key ingredients affecting tea quality, which is conducive to real-time monitoring of the current quality of tea leaves, evaluating the status during tea tree growth, and improving the quality of tea production. The application of this research will be helpful for the automatic management of tea plantations.

In the field of multi-robot cooperative localization and task planning, traditional filtering algorithms encounter synchronization and consistency issues during multi-source data fusion. These challenges result in cumulative localization errors and inefficient information sharing, which limits the system’s collaborative capabilities and control accuracy. To overcome these limitations, a distributed cooperative navigation strategy is introduced. Initially, a Distributed Adaptive Extended Kalman Filter (DAEKF) is implemented, which adaptively adjusts the noise covariance matrix to effectively manage nonlinearities and multi-source noise conditions. Subsequently, a Distributed Model Predictive Control (DMPC) framework is introduced. This framework predicts and optimizes each robot’s kinematic model, thereby improving the system’s collaborative operations and dynamic decision-making capabilities. Finally, the efficacy of this strategy is confirmed through detailed simulations and robotic experiments. The simulation results for cooperative localization demonstrate that DAEKF outperforms Kalman Filter (KF) and Extended Kalman Filter (EKF) in terms of localization accuracy. In the straight-line path-tracking experiments, DAEKF effectively reduced both lateral and heading errors for both robots. For Robot 1, DAEKF reduced the lateral error Root Mean Squared Error (RMSE) by 68.87%, 27.80%, and 25.76%, compared to No Filtering, KF, and EKF. In heading error, DAEKF reduced the RMSE by 52.29%, 41.89%, and 36.47%. For Robot 2, DAEKF reduced the lateral error RMSE by 51.30%, 22.88%, and 11.60%, compared to No Filtering, KF, and EKF. In heading error, DAEKF reduced the RMSE by 39.55%, 37.15%, and 26.00%. In the curved path-tracking experiments, both robots demonstrated high trajectory conformity while traveling along a predefined path combining straight-line and circular arc segments, with lateral errors in the straight-line segments all below 0.05 m. The strategy proposed in this study significantly enhanced the precision and stability of multi-robot collaborative navigation, demonstrating strong practicality and scalability.

Accurate classification of jujube varieties is essential for ensuring their quality and medicinal value. Traditional methods, relying on manual detection, are inefficient and fail to meet the demands of modern production and quality control. This study integrates hyperspectral imaging with intelligent optimization algorithms—Zebra Optimization Algorithm (ZOA), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Grey Wolf Optimization (GWO)—and a Support Vector Machine (SVM) model to classify jujube varieties. First, the Isolation Forest (IF) algorithm was employed to remove outliers from the spectral data. The data were then processed using Baseline correction, Multiplicative Scatter Correction (MSC), and Savitzky-Golay first derivative (SG1st) spectral preprocessing techniques, followed by feature enhancement with the Competitive Adaptive Reweighted Sampling (CARS) algorithm. A comparative analysis of the optimization algorithms in the SVM model revealed that SG1st preprocessing significantly boosted classification accuracy. Among the algorithms, GWO demonstrated the best global search ability and generalization performance, effectively enhancing classification accuracy. The GWO-SVM-SG1st model achieved the highest classification accuracy, with 94.641% on the prediction sets. This study showcases the potential of combining hyperspectral imaging with intelligent optimization algorithms, offering an effective solution for jujube variety classification.