Features of the implementation of an intelligent model for recognizing apple tree diseases by leaves
Abstract
Features of the implementation of an intelligent model for recognizing apple tree diseases by leaves
Incoming article date: 01.05.2025This article discusses the implementation features of a model for recognizing apple tree diseases by leaves. In the course of the work, a number of experiments were conducted with known convolutional network architectures (ResNet50, VGG16, and MobileNet). It was found that the accuracy decreases from ResNet50 to MobileNet, respectively. The effect of changing network parameters on accuracy is considered - the number of layers, batch normalization. A description of the practical experience of building our own model is given, which was also studied for changing parameters, such as the number of hidden layers. The network shows the best result with 3 and 4 layers. The nature of the learning ability of models from the number of layers and the range of epochs is studied. In early epochs, curves with a small number of layers show a rapid increase in accuracy, and curves with a large number are trained more slowly, which is probably due to the effect of gradient attenuation. In particular, it is shown that batch normalization or network deepening can positively affect such an effect as gradient attenuation. The use of batch normalization gave a pronounced effect of improving the network in the range from 35 to 50 epochs. For clarity, the work includes graphs of neural network training. In the process of work, conclusions and recommendations were formed on how to build a network more effectively.
Keywords: artificial intelligence, computer vision, neural networks, deep learning, machine learning