The major purpose of this study is to evaluate whether a Deep Learning (DL) model can perform better in image processing when it comes to identifying the presence of a disease in a pomegranate leaf.For this, a collection of 559 images from the Mendeley databases is gathered. There are images of both healthy and sick pomegranate leaves in this collection. Out of these, 540 images are chosen for testing, training, and validation of the DL models. However, the aforementioned operations can't start until the image has undergone preprocessing. The correct operation of the DL models is ensured by this preprocessing. Image resizing, image rescaling, and data augmentation are all included in the preprocessing.The steps of image resizing and rescaling make sure that the images are uniform in terms of size, dimensions, and pixel count. The data augmentation phase increases the training dataset's image count. This procedure extends the training period and ostensibly improves DL model performance. After preprocessing, the images are employed to train the built-in DL models. Three algorithms are used in the modeling process. The algorithms are Convolutional Neural Network (CNN), Convolutional Neural Network -Support Vector Machine (CNN-SVM), and Convolutional Neural Network-Long Short-Term Memory network(CNN-LSTM). All three models use the same number of images for training and validation. The models are evaluated for the final metrics after training and validation.Accuracy, True Positive Rate (TPR), True Negative Rate (TNR), False Positive Rate (FPR), and False Negative Rate (FNR) are some of the measures that were used in this study. Based on test results, a confusion matrix was also created. The CNN-LSTM algorithm is ultimately discovered to be able to offer a superior accuracy value than the other two methods. One final time, a real-time image is used to assess the model's performance. This study will likely be modified in the future to include the capability of both predicting and identifying the type of disease. Additionally, real-world images rather than those downloaded from the internet are anticipated to be utilized in place of the dataset used in this study to improve prediction accuracy.

Disease detection, deep learning models, image preprocessing, confusion matrix, etc.