Abstract

The application of the Convolutional Neural Network (CNN) algorithm through a website can provide assistance to tomato plant owners in the process of identifying the type of disease that attacks the plant. CNN is an architecture that consists of several stages and can be trained using images as input. The residual approach in ResNet makes it easy to learn the identity function by pushing the parameters in the weight layer closer to zero. These residual blocks can be used to train neural networks effectively by speeding up the flow of information through residual layer connections, as seen in the original ResNet with 152 layers. ResNet helps improve the accuracy of the CNN model and provides better predictive results in image processing and visual recognition. A detection system that uses the CNN algorithm and ResNet architecture will process tomato leaf images to accurately recognize disease symptoms and increase the accuracy of the detection model by overcoming the diminishing gradient problem in the CNN network. In terms of performance, ResNet shows the best results with a low loss value, namely 0.0435, and a high level of accuracy, reaching 98.63% on training data and 96.10% on validation data. GoogLeNet also produces good performance with a loss of 0.0677 and an accuracy of 98.29% on training data and 95.90% on validation data. However, it should be noted that although AlexNet has a higher loss, namely 0.4066, and lower accuracy, namely 89.63% on training data and 91.38% on validation data, this can be explained by differences in model structure.
Keywords: CNN, Tomato Leaves, ResNet