Quick detection of plant diseases and pests is of vital importance for preventing huge loses in agriculture and the environment by the hazardous of pesticide use and looks at the utilization of machine learning models, especially convolutional neural network (CNNs), for the detection and classification of plant diseases and pests. Different methods such as supervised as well as unsupervised learning jotted down. A unique CNN-LSTM + DENSENET201 hybrid model was developed by creating a deep feature extraction of pre-trained models such as DenseNet, ResNet, and GoogleNet with an LSTM ensemble classifier. The experimental studies on the plant datasets which included the images of various diseases of the crop showed the better accuracy and robustness of the hybrid model. CNN-LSTM + DenseNet201 model is one of the few that is over 99.4% accurate in real-time disease detection and outpaces other traditional and transfer learning-based models. By using unsupervised methods like anomaly detection and image restoration, the research avoids the need for high-quality labeled data sets when creating a cost-effective solution for the farmer. Further work will concentrate on the improvement of the model’s scalability along with the testing of its performance on additional datasets and plant types.