Identification of spectral signature of weeds and rice plants using hyperspectral sensor for early growth monitoring
2023
Hanzwan, M.N. | Juraimi, A.S. | Che' Ya, N.N. | Mat Su, A.S. | Ahmad A. | Nisfariza, N.M.
Rice (Oryza sativa L) is an essential staple food for more than 50% of the world's population. Unfortunately, weeds cause yield reduction by competing for light, water, and nutrients with rice plants. In addition, monitoring them visually in rice fields is challenging, especially for early growth detection. Hence, this study uses a hyperspectral sensor to identify the spectral signatures of grasses, sedges, broadleaved weeds, and rice plants for early growth detection. Hyperspectral images from four cultivated rice varieties, three weedy rice types, four sedge species, six grass species (excluding weedy rice), and six broadleaved weed species were acquired during early growth in a glasshouse condition. Then the mean spectrums for each species or variety were obtained using Spectronon Pro software before pre-processing for noise reduction and normalization. Orange data mining software we used for feature selection, classification model development, and model evaluation. The result showed that the spectral signatures for grasses, sedges, broadleaved weeds, and cultivated and weedy rice differ. Based on the relief-f ranking (feature selection), the top spectral bands for classification were in the near-infrared region. It was shown that the classification model predictions with feature selection achieved more than 90% recognition rates for grasses, sedges, broadleaved weeds, and cultivated and weedy rice plants. Furthermore the trained model with high classification accuracy has the potential to predict unseen new data to discriminate between rice and weed species in rice fields.
Mostrar más [+] Menos [-]Palabras clave de AGROVOC
Información bibliográfica
Este registro bibliográfico ha sido proporcionado por University of the Philippines at Los Baños