Recent developments in plant disease identification have considerably benefitted from the deployment of Convolutional Neural Networks (CNN). CNNs have demonstrated exceptional accuracy in recognizing different plant diseases, making them a preferred choice over traditional methods, which are often time-consuming and less effective. Ferrentinos (2018) provided a comprehensive analysis of deep learning models for plant disease detection, highlighting their ability to diagnose multiple diseases from image data with high precision [3]. This work opened up new avenues for investigation into the application of deep learning to agricultural operations. Building on this foundation, Shrestha et al. (2020) applied CNNs to plant disease detection, demonstrating the robustness of these models in accurately classifying various diseases. Their work, presented at the IEEE Applied Signal Processing Conference, showed promising results and underscored the potential of CNN architectures in enhancing plant pathology diagnostics [2]. Similarly, Deepalakshmi et al. (2021) created a CNN-based method for identifying illnesses in plant leaves, further confirming the effectiveness of CNNs in handling large-scale image data and providing reliable disease classification [4]. Several studies have conducted comparative analyses of different CNN architectures to identify the most effective models for plant disease detection. Sardogan, Tuncer, and Ozen (2018) explored a hybrid approach by combining CNNs with learning vector quantization. (LVQ) for plant leaf disease detection. Their findings indicated that the integration of LVQ with CNNs improved classification accuracy, showcasing the advantages of hybrid models in enhancing diagnostic performance [5]. Agarwal, Gupta, and Biswas (2020) focused on developing an efficient CNN model specifically for tomato crop disease identification.
Their model not only outperformed existing models in terms of accuracy but also demonstrated superior computational efficiency, making it a valuable tool for practical applications in agriculture [7]. The integration of machine learning models into mobile technologies has also been a significant focus in recent research. Wang et al. (2021) developed a trilinear CNN model (T-CNN) for the visual identification of plant illnesses, which was subsequently integrated into a mobile application for real-time diagnosis. This approach allowed farmers to use mobile devices for immediate disease detection, facilitating timely and effective disease management [8]. Joshi and Bhavsar (2023) suggested Night-CNN, a deep learning technology-based system designed mainly for mobile platform deployment, for the detection of nightshade crop leaf disease. Their model enabled real-time disease diagnosis in the field, emphasizing the practical benefits of mobile-ready diagnostic tools [9]. Further advancements have been made by incorporating additional techniques into CNN. models to improve their performance. Thakur, Sheorey, and Ojha (2023) proposed VGGICNN, a lightweight CNN model for crop disease identification. Their model achieved high accuracy while maintaining computational efficiency, making it suitable for use in resource-constrained environments [10]. Similarly, Lu, Tan, and Jiang (2021) reviewed various CNN applications in plant leaf disease classification, providing valuable insights into the strengths and limitations of different CNN architectures and their potential for improving agricultural practices [11]. Rao et al. (2022) developed a deep bilinear CNN for plant disease classification, which demonstrated significant improvements in classification accuracy by leveraging bilinear pooling techniques. This approach highlighted the potential of advanced CNN architectures in achieving higher diagnostic precision [12]. Suresh, Gnanaprakash, and Santhiya (2019) analyzed the execution of different CNN architectures with various optimizations for the categorization of plant diseases. Their study provided a comprehensive evaluation of CNN models, identifying optimal configurations for enhancing model performance [13]. In addition to these advancements, researchers have explored resilient CNN architectures to improve robustness against variations in image data. Gokulnath and Usha Devi (2021) developed a resilient LF-CNN for identifying and classifying plant diseases. Their model showed significant improvements in handling diverse image datasets, ensuring reliable disease detection under varying conditions [14]. Sun et al. (2022) conducted study on the diagnosis of plant diseases using CNN, further validating the efficacy of deep learning models in accurately diagnosing plant diseases [15]. The integration of hybrid models has also been explored to enhance disease detection accuracy. Singh et al. (2022) proposed a hybrid feature-based disease detection system that combined CNNs with Bayesian Optimized SVM and Random Forest classifiers. Their approach achieved high accuracy in plant leaf disease detection, demonstrating the benefits of hybrid models in leveraging the strengths of multiple machine learning techniques [16]. Ma et al. (2023) developed a YOLOv5n algorithm incorporating attention mechanisms for maize leaf disease identification. Their model showed significant improvements in detection accuracy, highlighting the potential of incorporating attention mechanisms in CNN architectures [17]. Lastly, evolutionary feature optimization has been explored to enhance the performance of deep learning models in plant disease detection. Al-bayati and Ustünda (2020) developed an optimization of evolutionary features technique for plant leaves disease detection using deep neural networks. Their approach demonstrated significant improvements in model performance by optimizing feature selection processes [18]. In order to detect diseases in plant leaves, Geetharamani and Arun Pandian (2019) used a nine-layer CNN. By using deep learning techniques, they were able to achieve high classification accuracy [19]. These studies collectively underscore the transformative potential of deep learning and mobile technologies in plant disease detection. The integration of CNN models into mobile applications represents a promising direction for real-time agricultural disease management, addressing critical challenges faced by farmers worldwide. As the field continues to evolve, further research is essential to enhance the robustness and scalability of these models, ensuring their widespread adoption and impact in agriculture. Our model's superior performance can be attributed to several key factors: Advanced CNN Architecture, Extensive Dataset, Data Augmentation and Robust Training, Real-time Application.
TABLE I: COMPARING ACCURACY & TECHNIQUE FROM RELATED WORKS.
Paper | Technique | Accuracy |
Singh et al. [16] | Bayesian Optimized SVM, and Random Forest | 96.1% |
Li Ma et al. [17] | YOLOv5 + Swin Transformer | 95.2% |
Arun Pandian et al. [19] | VGG16 | 92.87% |
Mihir Kawatra et al. [20] | AlexNet with GAP Layer | 97.29% |
Our Model | CNN | 98.14% |