研究目的
To explore the potential of close-range hyperspectral imaging with a snapshot hyperspectral sensor for identifying mangrove species under field conditions, and to determine the optimal spectral modes, relevant spectral wavebands, and effective classifiers for mangrove species identification.
研究成果
The study demonstrates that close-range snapshot hyperspectral imaging combined with machine learning classifiers, particularly SVM, is highly effective for mangrove species identification, achieving accuracies up to 96.46%. Derivative and logarithmic transformations, along with waveband selection, improve classification performance. This approach provides a practical tool for mangrove monitoring and conservation.
研究不足
Data acquisition was limited by accessibility, time, and illumination conditions, requiring supplementary data collection. The method is susceptible to weather and illumination variations, and involves time-consuming preparations and post-processing.
1：Experimental Design and Method Selection:
The study involved collecting field hyperspectral data of eight mangrove species using a snapshot hyperspectral imaging sensor, followed by data pre-processing, spectral transformations, waveband selection, and classification using machine learning algorithms.
2：Sample Selection and Data Sources:
Hyperspectral images were acquired from Qi’ao Island Mangrove Nature Reserve, with 60 sample spectra per species (total 480 samples) selected from healthy, sun-lit mangrove leaves.
3：List of Experimental Equipment and Materials:
UHD 185 hyperspectral snapshot sensor, notebook computer, standard white reference, dark measurements, and software including MATLAB R2014b, IBM SPSS Statistics 19, Weka
4：8, and Cube-Pilot. Experimental Procedures and Operational Workflow:
Data collection was performed on cloud-free days at solar noontime, with the sensor held approximately 20 cm above the canopy. Data pre-processing included radiometric correction and smoothing using the Savitzky-Golay algorithm. Spectral transformations (d(R), log(R), d[log(R)]) and vegetation indices were computed. Waveband selection methods (SDA, CFS, SPA) were applied, and classifiers (LDA, KNN, RF, SVM) were evaluated using 10-fold cross-validation.
5：Data Analysis Methods:
Accuracy assessment was done using confusion matrices, overall accuracy, kappa coefficient, producer's accuracy, and user's accuracy.
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
