Overall, agriculture's impact on a nation's economy is substantial, but in India it plays an especially vital role in improving the country's overall economic framework and standard of living. Crop yield in India is notoriously difficult to predict due to the country's erratic weather patterns and the prevalence of bacterial diseases. Researchers are focusing on this question in an effort to develop a method for identifying diseases in plants with small, green leaves at an early stage. Agriculture output relies on keeping plants healthy, and a single outbreak of a microbacterial infectious illness can wipe out an entire crop overnight. To this end, we set out to employ a deep-learning network-based technique for early detection of bacterial infections in green tiny leaves on green plants, before the onset of any outward signs. In this research, we evaluate recent advancements in the field of deep learning. To begin, we will investigate the deep neural network architecture, its data sources, and the various image processing methods that can be applied to the leaf images that have been recorded. In order to diagnose and categorize plant species, numerous DL architectures and data visualization's tools have emerged recently. To manage the functional automatic detection system for determining a particular plant illness in the field in which huge development of green small leaves on green organisms is mostly done, we had observed a problem that was unidentified in previous studies throughout our research survey and used their method to resolve that issue. CNN (convolution neural network) algorithms have recently undergone a number of improvements that result in more precise picture classification of any given object.To do this, we use the Agri-ImageNet collection in conjunction with realistically recorded photos of sunflower and cauliflower leaves, bulbs, and flowers. The PlantVillage dataset has regulated settings and uniform backdrops; this dataset fixes that. Deep transfer learning, which involves reusing knowledge representations, may alleviate these challenges. The primary goal of this research is to examine and evaluate all available deep transfer learning approaches in order to determine which one is most appropriate for a dataset consisting of plant species. In our study, we use a deep learning network to monitor plant symptoms in the leaf and, using the plant categorization information provided by the plant dataset, to determine the particular kind of bacterial infection at work. Our study illustrates the use of a Deep Learning algorithm implemented in SegCaps and EfficientNet Model to detect leaf diseases in green plants at an early stage and under varying environmental conditions. When compared to other illness detection methods using the HybridNet model, our findings demonstrate that SegCaps has an accuracy of 96.90%.