Classification of Elaeis Guineensis disease-leaf under uncontrolled illumination using RBF network

This paper proposes the novel technique of processing the leaf images from inconsistent illumination, as opposed to controlled environment approach used by conventional laboratory setup nowadays. The effect of daylight illumination creates major obstacle due to the changes of leaf appearance under different illumination that will affect the subsequent preprocessed image specifically when extracting the features such as spots and colours using RGB colour, histogram based texture and Gray Level Co-occurrence Matrices (GLCM). The key highlighted idea of processing the image under normal daylight involves colour manipulation technique to disclose the hidden RGB colour into its original colour, therefore the leaf images can be processed not limited to controlled environment setup. Once the illumination effect is alleviated, the oil palm leaf-diseased region can be automatically classified into three different types of disease namely nitrogen (N), potassium (K) and magnesium (Mg) deficiencies. The classification technique via Radial Basis Function Network (RBFN) classifier is examined and validated using an input from a database of oil palm diseased-leaf (or scientifically known as Elaeis Guineensis) to confirm the high degree of classifier capability to recognize the disease according to the deficiency type. The experimental result showed that various lighting conditions measured in terms of illumination affected the formed pattern on the leaf surface due to the leaf images dependency on the intensity of illumination presented at that time hence disturbing the image quality. Classification of the various deficiencies demonstrated the correct classification rate as high as 70.69%, showing that the inconsistent illumination could provide good accuracy in classifying the deficient-leaf.