Study on Venation Visualization System Using Hyperspectral Images and Multi-Layer Perceptron Classifier

Commonly, a leaf venation is visualized using an image acquired from an RGB camera in the visible light spectrum and processed morphologically. In this paper, we propose a novel method for visualizing of the leaf venation. The proposed method consists of a hyperspectral camera on the Visual-Near Infrared band as image acquisition system and a Multi-Layer Perceptron Classifier (MLPC) as a classification algorithm. In this study, we compare some activation functions and optimizers to find the proper classification model for leaf venation. The Red Amaranth leaf was used as a sample that acquired using the hyperspectral camera at band 400-1000 nm. We choose two classes to represent the leaf part namely a vein area and non-vein area. The five-square pixels in the leaf image were used to represent the vein and non-vein object. The averaging of the spatial area at the full band was conducted as a spectral feature of the object. Five-fold cross-validation was performed to evaluate the performance of the proposed method. Accuracy, precision, and recall scores were computed for each classification model. The best classification result has accuracy 94.9% using activation function linear and solver function of Limited-memory Broyden-Fletcher-Goldfarb-Shanno (lbfgs). The best model is then used for visualizing venation using the hyperspectral image. The result shows that the best model could visualize primary and secondary veins in the leaf. Thus, the proposed system can be used for visualizing leaf venation on Red Amaranth leaf.