Majority voting transfer learning CNN for peanut leaf types identification

Abstract

In agricultural information systems, an effective way to identify plants is from their leaves because they have different characteristics and types, are easy, do not damage trees, and do not even need to be picked. Research related to the introduction of plant species using leaf-based CNN (Resnet101, ResNet50, ResNet18, MobileNet V2, DenseNet201, and GoogleNet) transfer learning has been carried out previously, but it is still not effective because of how to recognize peanut plant species using a single leaf and the accuracy value an average of 82.97. So this study proposes Majority Voting from the identification of the CNN transfer learning method to be able to effectively identify the type of bean based on the leaves of all the leaves on the stalk. The Majority Voting technique proposed is based on the type of class that is the most majority or dominant. Selection the proposed majority experimented with datasets of peanut leaf types, namely mung bean, soybean, long bean, and peanut leaves, a total dataset of 456 images of peanut leaves. Data collection is done by taking directly on the farmer’s land. The CNN transfer learning model used is Resnet101, ResNet50, ResNet18, MobileNet V2, DenseNet201, and GoogleNet. Results the Majority voting of proposed transfer learning has an accuracy of 96.93.