Deep learning, especially Convolutional Neural Networks (CNNs), has become a leading technique for image recognition, excelling in fields like disease diagnosis, facial recognition, and traffic sign detection. Their success has encouraged research in agriculture, where CNNs are applied to tasks such as plant species identification, yield optimization, weed detection, and disease recognition from leaf images. Selecting the best CNN model for specific agricultural datasets is challenging due to the wide variety of research. This survey reviews CNN applications for detecting and classifying soybean leaf diseases. It compares pre-processing approaches, CNN architectures, frameworks, and optimization methods used on leaf image datasets. The paper also evaluates performance metrics and datasets commonly used to assess model accuracy. By analysing strengths and weaknesses of different techniques and models, the survey aims to guide researchers developing deep learning solutions for plant disease prediction, identification, and classification.