Olive trees are essential in many world regions, providing valuable products such as olive oil and table olives. However, olive trees are susceptible to several diseases that can significantly affect their health and productivity ^[1]. Early and accurate diagnosis of these diseases is critical for effective management and control ^[2]. Artificial intelligence (AI) finds application in a wide range of domains, playing a pivotal role in addressing various challenges across industries such as ^[3], agricultural ^[4], medical ^[5], and more during their early developmental stages. Using AI with outdated or incomplete data can limit the accuracy and relevance of models. This effect hinders the smart applications’ ability to make informed decisions or predictions. Enhanced images are necessary before use if the application works on a computer vision area. Image enhancement based on balanced colors and brightness finds applications in various fields, such as photography, medical imaging, satellite imaging, surveillance, and computer vision tasks ^[6]. In computer vision tasks such as object detection or recognition, enhancing images beforehand can improve the performance of the algorithms by making the visual features more distinguishable ^[7].

In the proposed model, AI is harnessed to integrate clustering and deep learning, facilitating the identification of olive diseases through a comprehensive analysis of olive leaf images. Figure 1 shows the implementation scenario that was used in the proposed model.

The proposed scenario uses IoT technology or computer vision tools within an olive farm to collect images from sensors and imagery. The goal is to improve disease diagnosis and general monitoring of olive trees. By integrating intelligent technologies, farmers can proactively manage crop health and promptly address potential issues, contributing to more sustainable and productive farming practices. The proposed model demonstrates a robust and practical approach to olive disease diagnosis, combining image enhancement techniques and deep learning-based classification to achieve accurate and reliable results.

The motivations for this model lie in addressing the challenges faced by the olive industry, leveraging AI and computer vision techniques to enhance disease detection, and providing a comprehensive and effective solution for olive disease diagnosis. Olive trees are essential in many regions, providing valuable products such as olive oil and table olives. Ensuring the health and productivity of olive trees is critical to sustaining these industries and the livelihoods of the people who depend on them. Olive trees are susceptible to several diseases that can significantly impact their health and productivity. Olive groves are exposed to various pathogens, including Aculus olearius, which can make developing an effective disease detection algorithm as challenging as creating a universal classification model [4,8]^[4][8]. Early and accurate diagnosis of these diseases is critical for effective management and control to prevent widespread damage. The current deep learning models used in computer vision have several drawbacks:

• Safer from unbalanced data: It can significantly affect the performance of deep learning models, leading to biased predictions and lower accuracy. A lack of sampling of minority classes can cause the model to prioritize the majority class and neglect essential patterns in the data [9,10]^[9][10]. Techniques such as resampling, synthetic data generation, and special loss functions are commonly used to mitigate the adverse effects of data imbalance in deep learning.
• Low accuracy with noise data: Noisy data or low-visibility images can significantly impact the accuracy of deep learning models ^[6]. Noise can distort underlying data patterns, leading to misclassifications and reduced performance. To address this issue, robust preprocessing techniques, noise reduction methods, and regularization strategies may enhance a CNN’s ability to extract meaningful features from noisy input.
• High complexity: The size of samples in deep learning contributes to increased complexity, impacting training times and risking overfitting. Techniques such as dimensionality reduction help mitigate this challenge by improving efficiency and generalization.

2. Olive Leaf Disease Diagnosis

The development of machine learning models saw numerous studies proposing different approaches for classification and regression tasks. However, many of these research endeavors primarily relied on standard machine learning or deep learning classification algorithms, often without adequately addressing the limitations associated with these models. It is worth noting that machine learning algorithms tend to face challenges that could be addressed with better-prepared data, higher accuracy rates, and reduced complexity in terms of space and time. Alshammari et al. ^[4] proposed a deep learning model combining vision transformer and convolutional neural network architectures for accurate olive disease detection. The study combines the performance of two different deep learning architectures, namely vision transformer (ViT) and a convolutional neural network (CNN), to improve the accuracy and efficiency of disease detection in olive plants. The weaknesses of the proposed system lie in improper training, as the model was trained on imbalanced data, which may affect its learning ability and exacerbate other problems related to uneven class distribution or accurate class prediction for classes with small counts. In ^[11], the study focuses on using the capabilities of convolutional neural networks (CNNs) to identify and classify different diseases affecting olive leaves accurately. By applying deep CNNs to the problem of olive leaf disease classification, the work is consistent with the trend of using advanced machine-learning techniques to improve disease detection and monitoring. Without sufficient emphasis on regularization techniques, hyperparameter tuning, or cross-validation, the proposed model may be susceptible to overfitting the training data. Overfitting could lead to poor performance on unseen data and undermine the practical utility of the model. Ksibi et al. ^[12] proposed a hybrid deep learning model for olive leaf disease detection and classification. The proposed model is composed of neural networks from the ResNet50 and Mo-bileNet models. The model combines elements of the MobileNet and ResNet architectures to improve its ability to identify and categorize different diseases affecting olive leaves accurately. This approach aims to leverage the strengths of the different architectures while mitigating their weaknesses. However, its computational cost and memory requirements may limit its applicability in resource-constrained environments. In ^[13], the author used the deep learning architecture of Inception V3 to accurately and efficiently classify olive leaf diseases. This framework addresses the critical challenge of identifying diseases affecting olive leaves by leveraging the capabilities of a robust convolutional network architecture. By leveraging the strengths of Inception V3, the framework aims to achieve higher accuracy in distinguishing between different types of olive leaf diseases. Despite the complexity of the framework, the model can achieve a high level of accuracy. Gulzar ^[14] presents a study on using deep learning techniques for fruit classification. He used the MobileNetV2 with the deep transfer learning technique to classify fruit images. The classification layer of MobileNetV2 is replaced by a customized head, which produces the modified version of MobileNetV2 called TL-MobileNetV2. In addition, transfer learning is used to retain the pre-trained model. The study findings indicate that transfer learning significantly contributes to better results, and the use of the dropout technique helps mitigate overfitting in transfer learning. Mamat et al. ^[15] proposed a deep learning model to enhance the classifying of the ripeness of oil palm fruit and recognize a variety of fruits. The authors proposed simple and effective models using a deep learning approach with You Only Look Once (YO-LO) versions. There are several issues with the proposed model that need to be considered. These include difficulties in dealing with overlapping objects, detection of partially hidden objects, and possible biases in the data used to train the model. In addition, the performance of the proposed model may be affected in certain situations due to the balance between spatial resolution and speed and its complicated structure.