A Deep Learning-based Approach for Edible, Inedible and Poisonous Mushroom Classification

Abstract

Mushroom is one of the fungi types' foods that has the most powerful nutrients on the plant. However, there are as yet different mushroom species that have not been identified. Nevertheless, the identification of edible, non-edible, and poisonous mushrooms among its existing species is a must due to its high demand for peoples’ everyday meals and major advantage on medical science. For this purpose, we applied InceptionV3, VGG16, and Resnet50 on 8190 mushrooms images where the ratio of training and testing data was 8:2. Contrast Limited Adaptive Histogram Equalization (CLAHE) method was used along with InceptionV3 to obtain the highest test accuracy. CLAHE is a variant of Adaptive Histogram Equalization (AHE) which improves the local contrast-enhancing of images and the definitions of edges in each area of an image by limiting over amplify noise is moderately similar regions of an image. We show a comparison between contrast-enhanced and without contrast-enhanced methods. Finally, InceptionV3 has achieved 88.40% accuracy which is the highest among the rest implemented algorithms.