An intelligent agricultural system is considered a crucial aspect of the modern world. Deep learning is the most popular and extensively tested approach to developing artificial intelligence. The development of agriculture through the application of deep learning techniques is a topic of significant interest. Deep learning models, especially those based on convolutional neural networks (CNNs), have been employed for precision agriculture. In our study, we investigate the impact of six distinct augmentation techniques on a limited dataset. We manually gathered data from the agricultural plant insect dataset (BAU-Insectv2) and employed four state-of-the-art CNN architectures (ResNet-101-v2, EfficientNet-B1, Inception v3 and Inception- ResNet v1 ) with various optimizers and augmentation techniques. We observed how these factorsinfluenced the dataset in our image classification experiments. Among the three optimizers (ADAM, SGD, RMS-prop), we achieved the best performance and highest accuracy with ADAM on a small dataset. We achieved superior image classification results when applying a minimum of four different augmentations and using an ADAM optimizer. Four CNN-based models are employed to enhance performance measures. The research focused on evaluating performance using various metrics, including accuracy, sensitivity, specificity, precision, recall, and F1- score. The primary aim of this study is to continually enhance the model’s performance over time. Finally, we have presented a comprehensive framework designed for real-time crop image classification which includes the development of a user-friendly web application for seamlessand efficient operation.