Significant impacts on agricultural productivity come from peanut leaf diseases. Recurrent advances in deep learning have facilitated the development of models to detect diseases with good accuracy. This research work presents a detailed observation of using deep learning models for the identification of peanut leaf diseases, focusing on two different architectures: DenseNet 169 and Inception V3. A detailed analysis of the performance of two different architectures has been done to detect diseases in peanut leaves. Initially, the manuscript leverages trained models and enhanced the detection accuracy while decreasing the computational costs by using transfer learning. Then it evaluates image preprocessing techniques for quality and robust improvements of the data set. Finally, it analyzes the filtered convoluted images from DenseNet Architecture’s dense blocks and compares model performance criteria like as F1-score, accuracy, precision and recall. The study demonstrates the efficacy of transfer learning and picture preprocessing approaches such as augmentation in improving model accuracy and precision. The models can detect six classes of diseases, i.e., Healthy Leaf, Early Rust, Late Leaf Spot, Rust, Early Leaf Spot, and Nutritional Deficiency. Key findings of the study include the superior performance of DenseNet-169 with RMSProp optimizer and learning rate = 0.0001, achieving an accuracy of 98.42% and precision of 98.47% over Inception V3 with RMSProp optimizer and learning rate = 0.0001, achieving an accuracy of 95.70% and precision of 95.48%.