Humans' ability to extract information from images is more accessible than machines. The ability of human vision is extraordinary because they have little or no supervision when recognizing objects regardless of the similarity of images. Early studies of visual recognition have shown that machines perform better than humans when there is enough information for prediction and classification. This is less efficient for machines. In this paper, we propose a new way to solve this problem using the provided plant dataset, which will use visualization techniques to solve the problem when the model finds itself in a limited data scenario. Our approach yields more promising results than state-of-the-art models. We used three different types of datasets, including benchmarking Plant Village and Plant Doc. These datasets have controlled, uncontrolled, and downloaded images from the internet. Each dataset is used for our model, resulting in better performance than state-of-the-art results.