研究目的
To develop a rapid and nondestructive method for visually detecting the moisture content of tea leaves, addressing the issue of spectral differences between the front and back sides of leaves in a production-like environment.
研究成果
The research successfully proposed and validated a scheme for automatic visual detection of tea leaf moisture content, addressing spectral differences between leaf sides. LS-SVR models showed high accuracy, and a logistic classifier achieved perfect side identification. The method is feasible for practical applications but requires further optimization for real-world conditions.
研究不足
The study was conducted in a lab setting, which may not fully replicate the complexities of actual production environments, such as various conveyor belt types, leaf curvatures, shadows from overlapping leaves, changes in detection distance, illumination uniformity, and random leaf postures.
1：Experimental Design and Method Selection:
The study used hyperspectral imaging technology to capture images of tea leaves, simulating a practical production environment. Methods included hyperspectral calibration, feature band selection using Random Frog and Successive Projection algorithms, and regression modeling with PLSR and LS-SVR. A logistic regression classifier was designed for side identification.
2：Sample Selection and Data Sources:
'Wuniuzao' tea leaves were collected from an experimental garden, with 64 leaves used. Hyperspectral images were acquired for both sides of the leaves during multiple drying cycles.
3：List of Experimental Equipment and Materials:
Hyperspectral image acquisition system (spectrometer, lens, light source, conveyor system, computer, camera obscura), electronic balance, oven, microscope, scanning electron microscope.
4：Experimental Procedures and Operational Workflow:
Leaves were weighed, scanned, dried in an oven, and rescanned over cycles. Data processing included calibration, average spectrum extraction, sample division, abnormal sample detection, feature selection, model building, classification, and visualization.
5：Data Analysis Methods:
Statistical analysis included RMSEC, RMSEP, RC2, RP2 for model evaluation, Monte Carlo sampling for abnormality detection, and logistic regression for classification.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
