Diagnosis of potato leaf diseases is an important factor in ensuring the sustainability of agriculture. The paper introduces a hybrid deep learning approach (PhytoResLSTM) to classify potato leaf diseases based on ResNet50 and BiLSTM architectures. In this approach, ResNet50 was used for deep feature extraction and bidirectional long short-term memory (BiLSTM) network to capture the long-term relationships of sequential data to improve disease classification accuracy. The proposed method was tested using PlantVillage dataset that containing three classes: late blight, early blight and healthy leaf images. The metrics used in evaluation such as accuracy, precision, recall and F1-score show that the model achieved 98.61 % accuracy in classifying leaf disease. The proposed technique with precision of 94.62 % and recall of 97.96 % validates early detection of the leaf diseases, where false positives and false negatives are reduced to a minimum. An F1-score of 96.15 % represents an ideal balance between accuracy and error reduction. The model perfectly identified the early and late blight with the area under the curve (AUC) of 1.00 and 0.97 for healthy leaf with an overall ROC-AUC of 0.98 exhibiting excellent classification. The confusion matrix further confirms that the model demonstrates excellent performance, characterised by minimal misclassification in the identification of the potato leaf diseases. According to the state-of-the-art comparison the proposed hybrid model outperforms conventional feature extraction and leaf disease classification methods underscoring the strength of hybrid deep learning methods and suggesting their application to improve food security and sustainable agricultural systems.