Bangladeshi Plant Recognition using Deep Learning based Leaf Classification

At present deep learning-based object recognition approaches have placed a tremendous effect for classifying different objects. Leaves recognition using supervised learning has shown satisfying performance which may help in various research purposes also. In our work, we have used a deep convolutional neural network as a classifier. We have used a transfer learning approach. We have prepared our work dataset based on Bangladeshi plants which contains eight different classes of leaves. We have experimented with VGG16, VGG19, Resnet50, InceptionV3, Inception-Resnetv2 and Xception deep convolutional neural network models where we have found the highest value in VGG 16 which shows almost 96% classification accuracy. Recognition of useful plants using leaf image will be greatly helpful in the research of ayurvedic and endangered plants.