The daylily (Hemerocallis citrina Baroni) is an herbaceous perennial whose flowers are rich in nutritional and functional components. It is typically cultivated in rows but harvested manually, a labor-intensive process that this study aims to automate by developing a robotic harvesting system. A critical component of such a system is autonomous navigation, which poses significant challenges in unstructured field environments due to changing natural light, randomly distributed weeds, and varying inter-row density. To address these challenges, this study adopted vision-based navigation technology and proposed a guidance directrix detection algorithm. The proposed approach begins with converting the color model from RGB to HSV to decouple the brightness, thereby mitigating the impact of natural light variations. Morphological dilation is applied to suppress noise from weeds in the inter-row regions. Furthermore, an innovative coarse segmentation strategy based on hue and saturation is introduced to handle the problem of different sparsity in the inter-row. Finally, the inter-row axis is accurately extracted by employing concepts from physics, namely, the center of mass and moment of inertia. Experimental results demonstrate an accuracy of 98.25%, a recall of 100%, an average navigation processing time of 8.7521 ms, and a compact model size of only 18 KB. These findings empirically confirm that the proposed approach achieves high precision and real-time performance under unstructured, uncertain, and dynamically changing field conditions. Additionally, the algorithm operates with high computational efficiency and requires neither expensive hardware nor large-scale training datasets.

For large-scale rapid detection of the gelatinization parameters of millet flour, a method to predict the gelatinization characteristics of millet flour was explored using hyperspectral imaging combined with deep learning. The average spectral data of millet flour were obtained through successive hyperspectral data feature extraction and preprocessing, and based on the data matrix obtained, a regression model to predict the gelatinization parameters of millet flour samples was developed using a back propagation (BP) neural network optimized by sparrow search algorithm (SSA). The results showed that the spectral data pre-processing program used in this study could standardize and simplify the process of spectral data extraction and pre-processing, and this program was generally applicable to spectral data extraction and pre-processing for powder and fine particle samples. BP algorithm and SSA-optimized BP algorithm were used to predict the gelatinization parameters of millet flour. The mean square error (MSE) between the prediction value and the tested value of each parameter decreased after optimization of BP algorithm, from 0.0266 to 0.0175 for peak viscosity. Therefore, the SSA optimized BP algorithm could predict the gelatinization properties of millet flour more accurately. This study can provide theoretical support for the application of hyperspectral imaging coupled with deep learning in the prediction of the gelatinization properties of millet flour.