Analysis and Development of Eight Deep Learning Architectures for the Classification of Mushrooms

Abstract

One food item that is easy to find in nature is the mushroom. In terms of form and characteristics, mushrooms are similar. Arranging mushrooms into groups so that poisonous and non-poisonous ones can be separated is important. Real-time analysis of mushrooms is still not used very often. Previous studies focused primarily on performance and accuracy, ignoring architectural computing and a significant amount of data preprocessing. The data set used is more laboratory-conditioned. This will impede the process of widespread implementation. The study suggests changes to eight current architectures: Modified DenseNet201, DenseNet121, VGG16, VGG19, ResNet50, InceptionNetV3, MobileNet, and EfficientNet B1. The development of this architecture took place within the areas of classification and hyperparameter learning. In contrast to the other eight architectures, the MobileNet architecture exhibits the lowest computational performance and highest accuracy, according to the comparison results. When the confusion matrix is used for evaluation, an accuracy of 82.7% is achieved. Modified MobileNet has the best speed because it keeps a lower computation architecture and cuts down on unnecessary preprocessing. This means that many people can use smartphones with more realistic data conditions to make it work.

Keywords

• mushroom
• deep learning
• modified mobilenet
• classification