Hyperspectral Data Compression Model Using SPCA (Segmented Principal Component Analysis) and Classification of Rice Crop Varieties

Abstract

Hyperspectral Image Processing System (HIPS) is a good source of vegetation detection and identification. This work presents a spectral classification of rice crop using EO-1 Hyperion Hyperspectral image. In HIPS the traditional classification methods have major limitations due to high dimensionality. The Principal Component Analysis (PCA) is a well established data compression tool that can be applied on Hyperspectral data to reduce its dimensionality for feature extraction and classification. Now PCA has become a traditional tool of data compression in HIPS. This research proposes a new approach of data compression based on Segmented Principal Component Analysis (SPCA). The outcomes of our analysis led to a conclusion that the SAM classification of PCA _NIR (671.02-925.41nm) discriminates RICE crop varieties RICE 1[Ratan (IET-1411)], RICE 2[CSR-10 (IET-10349/10694)], RICE 3[Haryana Basmati-1(IET-10367)], RICE 4[HKR-126] and RICE 5[CSR-13 (IET-10348)] better than traditional PCA _{VNIR − SWIR} and PCA _VIR , PCA _{SWIR − 1}, PCA _{SWIR − 2}, PCA _{SWIR − 3} segments. Results of this research work have shown that the overall classification accuracy of PCA₅ in PCA _NIR segment is achieved 80.24% with kappa coefficient 0.77, however RICE4 and RICE5 varieties are classified 100% and RICE1 (72.73%), RICE2 (85.71%) and RICE3 (91.67%) are classified more accurately than other classification results.