<p>Plants face various stresses during their growth phase. If these stresses are not detected early, they can adversely impact plant growth and yield. Therefore, the main objective in enhancing productivity is early identification and characterization of these stresses. In the agricultural sector, deep learning (DL) has been employed to solve numerous issues, including those related to diseases. Consequently, a variety of AI-driven techniques have emerged for the detection of plant leaf diseases. However, a significant challenge remains in establishing an effective early disease detection method that has reliable evaluation metrics. This study presents an efficient and successful solution named AutoDL (Auto-encoder and CNN), which merges the strengths of an optimized Convolutional Neural Network (CNN) with those of an autoencoder. The research uses a publicly available tomato dataset comprising 10,000 leaf images categorized into ten classes to fine-tune AutoDL. First, the autoencoder is applied to extract essential features from the images for the purpose of classifying and detecting leaf diseases. These extracted features are subsequently used to refine the weights of the CNN. With 10,000 training parameters, AutoDL achieves a disease detection accuracy of 98.754%, showcasing remarkable performance compared to other methods. In this context, the study employs two deep networks, adjusting the weights of the features obtained from the input images, which leads to improved results compared to existing state-of-the-art models.</p>