Onions are popular because of their rich in nutrients such as protein, vitamin C, and phenolic compounds, which help guard our bodies against serious health problems like heart conditions, osteoporosis, and high blood sugar levels. They’re also crucial for keeping our digestive system in good shape. Due to the broad use of onions in areas like health, farming, and our daily routine, it’s critical to pinpoint and deal with diseases that affect them. Farmers often struggle to identify leaf diseases at an early stage. Consequently, the disease spreads badly in the crop field, and farmers face a loss in the cultivation season. We, therefore, need an automated onion leaf disease detection system that can diagnose diseases effectively. In this study, we developed a deep learning model that includes soft attention and Long Short-Term Memory (LSTM) features. We enhanced 3,660 images of four classes, including Botrytis Blight (BB), Downy Mildew (DM), Purple Blotch (PB), and Stemphylium Leaf Blight (SLB) using gamma correction and a green fire-blue filter, intending to enhance the disease spots and boost the performance of our models. We also employed several advanced learning methods to compare with our proposed model: VGG16, VGG19, MobailnetV2, ResNet50, InceptionV3, and InceptionResNetV2. However, our proposed model achieved the best accuracy rate of 96.38%. We tried make it the most accurate and least error-prone compared to the other methods. So, it can be concluded that our proposed approach will help farmers and specialists to diagnose leaf diseases.