From Pixels to Insights: Transforming Plant Pathology with Artificial Intelligence

We urgently want more precise methods of detecting plant diseases because they pose an increasing danger to the global food supply. Finding a solution to this challenge utilising state-of-the-art deep learning technology is the focus of this research. Various models' abilities to diagnose plant diseases are examined in the study. Squeezenet1_1, VGG13, Inception_v3, ResNet34, DenseNet169, and AlexNet are among the models it examines. Included in the data set is a background class from the DAGS dataset in addition to 38 distinct plant diseases culled from the PlantVillage dataset on Kaggle. The results demonstrate that various designs and training methods exhibit slight variations in performance. For instance, after extensive retraining, AlexNet achieves an accuracy of 99.24%. However, after extensive retraining, DenseNet169 achieves an impressive 99.72% accuracy. In order to have a deeper grasp of the models' behaviour, the research incorporates visualisation experiments with accuracy measurements. In the occlusion experiment, we can see heat maps that graphically depict how various picture elements impact the program's decision-making. In order to determine which pixels are most relevant, the saliency map experiment use mathematics. This sheds light on the decision-making process of the model. Utilising AlexNet and VGG13, pictures of specific plant diseases such as Septoria Leaf Spot, Early Blight, and Late Blight are displayed. To summarise, this article discusses the inner workings of various deep learning models for plant disease detection and compares them. The findings aid in selecting optimal models and training strategies, which in turn facilitates the development of robust and reliable automated systems for disease diagnosis in plants. Using cutting-edge technologies, this study represents a major step towards improving agricultural disease detection.