부산대학교 도서관

Convolutional neural networks (CNNs) have had remarkable success in classifying a variety of plant diseases through deep learning. However, just a few researches have shed light on the inference process, leaving it as an unsolvable mystery. In addition to guaranteeing the learnt feature's dependability, revealing the CNN to extract it in an understandable form enables human intervention-based verification of the model's veracity and the training dataset. Using a CNN that had been trained using a public ally accessible collection of images depicting plant diseases, several neuron-wise and layer-wise visualisation techniques were used in this study. We demonstrated that neural networks can, when diagnosing an illness, capture the hues and textures of lesions particular to that disease, which is similar to human judgement. The most critical aspect of agriculture is striking a balance between produce and population. Due to a variety of issues, including natural disasters, unforeseen rainfall, nutrient deficiencies in the soil, etc., the majority of farmers failed to produce and balance the crops. The main issue, however, is pest infection, which is the root of all the issues. To learn about plant diseases, several researchers employed a variety of methods. Convolutional neural network-based deep learning techniques are frequently utilised to solve image-oriented problems. A powerful and successful method for image analysis is the CNN (ConvNet) neural network model of deep learning. In this study, various models for plant disease detection with CNN are compared. The research report concludes by summarising its findings, identifying its limitations, and making recommendations for classification