Wavelength Selection and Classification of Hyperspectral Non-Imagery Data to Discriminate Healthy and Unhealthy Vegetable Leaves

Being the largest vegetarian population across the globe, vegetables are an integral part of Indian meals. The proposed research finds significant wavelengths to discriminate healthy and unhealthy vegetable plants. Spectral-reflectance (SR) and first-derivative (FD) in the visible, red edge and near infrared region (350–1000 nm) of three vegetables brinjal, cluster beans and long beans were used. The significant wavelengths were selected using ReliefF and Support- Vector-Machine (SVM). Random forest algorithm was used for classification. The binary classification was used for each vegetable separately, and multiclass classification was applied for all the samples. The most significant spectral wavelengths, for the prediction of diseased brinjal, correspond primarily to the red edge in SR. Long beans samples were classified accurately in the red-edge. In the case of cluster beans, SR is more effective than FD in the red-edge. The results substantiate the utility of HS data for discrimination of healthy and unhealthy vegetable plants and even vegetable types.