Development of deep learning models for predicting plant diseases in agriculture based on wavelet transformations

Abstract

Using wavelet transformations and deep learning methods, models for the classification of plant diseases (PlantVillage) have been built. CNN models with wavelet transformations for plant classification are constructed. Models of classification of diseases of plant leaves, which are crucial for effective crop protection, are constructed. Effective methods have been developed and CNN convolutional neural networks have been built. The main achievement of the work is the integration of deep learning methods with WaveletFusion transformations. A new approach to deep learning modeling uses various types of special activation functions to improve model accuracy. The proposed technology integrates channel-by-channel concatenation of stationary wavelet transform (SWT), discrete wavelet transform (DWT) and grayscale images in the encoder-decoder architecture based on transfer learning of neural modeling.