Multi-class Guava Disease Classification using an Efficient and Fine-Tuned DenseNet model

Growing guavas is important for farming in many places, but due to the various guava diseases, the yield may get reduced. In the traditional scenario, it becomes very difficult to detect and reduce the effect of diseases on the guava fruit. This study looks into using state-of-the-art deep learning methods, especially the DenseNet model. It helps in the early detection of guava disease quickly and with good performance outcomes. The new model uses a DenseNet design with Adam and SGD optimizers. It reaches high accuracies of 98.76% for the Adam one and 97.3% for the SGD optimizer, after about the 30th epoch the performance of the SGD optimizer started to degrade, whereas, the highest and best performance with Adam was achieved at epoch 43. The study looks at how long it takes to train a model and its performance. It finds that training for 43 cycles gives the best balance of time savings and quality results. When the comparison of both the optimizers has been done, we found that the Adam outperforms and results in the F1-Scores of 0.9612 and 0.9523 for Adam and SGD optimizers, respectively. This shows that the model can tell apart different types of diseases in guava leaves with effective and efficient outcomes. The proposed model can also sort out different types of leaf problems and may help to classify illnesses in other crops as well with more training data. This study helps us understand how deep learning, like the DenseNet model, can be used to rightly predict diseases in guava crops. This is important for eco-friendly farming.