Prediction of Disease in Apple Leaf Using Convolution Neural Networks

Abstract

Apple tree diseases can significantly impact fruit quality and yield, making early detection and intervention crucial for orchard management. To avoid the impact on apple production a novel approach is required for the early and accurate prediction of apple leaf diseases through Convolutional Neural Networks (CNNs). Our proposed system contains a comprehensive dataset of high-resolution pictures of apple leaves exhibiting various disease symptoms, including common issues like apple scab, apple rust, and powdery mildew. The dataset was carefully annotated to train and validate the CNN model effectively. The proposed CNN model makes use of its ability to automatically learn applicable features from images, making it highly suitable for the task of disease prediction in apple leaves. This model contributes to the field of precision agriculture by offering a cost-effective and efficient tool for apple disease detection.

Introduction

I. INTRODUCTION 

In the last few years, the agricultural sector has witnessed a surge in the utilization of advanced technologies, particularly in disease identification and management. Among various crops, apple cultivation holds significant economic importance worldwide. However, one of the most difficult tasks for apple growers is the early detection and control of diseases that can have a noteworthy effect on yield and quality. Apple is among the globally consumed fruits among the four after bananas, grapes, and oranges. Globally, using pests and fertilizers can affect the apple plant and also the standard of fruit which should show an influence on the overall production of the apple crop In India, fungal diseases are the major ones that affect the apple fruit quality in particular apple producer state i.e. Himachal Pradesh, which is the second-biggest apple producer in India.

Infections in plants include diseases such as biotic and abiotic. Biotic diseases like scab, cedar rust, leafy blotch, powdery mildew, blight, mosaic, black rot, etc are very dangerous compared to abiotic diseases. It is crucial to increase apple production and early identification of disease. There are some serological method-based techniques for disease identification such as Fluorescence in-situ hybridization (FISH), polymer chain reaction (PCR), immunofluorescence (IF), and flow cytometry used by experts and can only be performed in a laboratory.

Machine learning-based approaches such as K-nearest Neighbor (KNN), Artificial Neural Networks (ANN), and Support Vector Machine (SVM) are mostly employed for disease identification. Convolution Neural Networks (CNN), one type of deep learning technique, have efficiency and it has also proved their suitability for plant disease identification. A deep CNN model consisting of three convolution layers is developed for disease detection based on apple leaf images. This model can self-learn hierarchical features, making it ideal for tasks such as image categorization and object detection. The objective is to create a robust and accurate a model capable of identifying common diseases affecting apple trees such as scab, rust (or) black rot.

II. LITERARATURE REVIEW

A. Transfer Learning Approach

[1] In this research study, a transfer learning strategy was employed to identify apple leaf disease using the CNN model. Transfer learning is a method of using a pre-trained model as a starting point for a new task. This study shows that a transfer learning strategy can accomplish high detection of apple leaf disease using a relatively small dataset (consisting of 2897 images) of training data. 

First, you need to choose a pre-trained CNN model trained on a large dataset such as image-Net. At the top of the trained network, remove the fully linked layer. Freeze the convolutional layer weights if necessary. This prevents updates during training and preserves previously learned features. Add a new fully connected layer above the convolutional base. Train the model using the collected dataset. Experimentation results on a gathered dataset of 2897 images with data augmentation demonstrated that  AppleNet can be effectively used to detect apple diseases on apple leaves with a classification accuracy of  96.00%. 

B. Mobilenet CNN Model

[2] The study proposed a MobileNet CNN model for identifying disease in apple leaves.MobileNet is a lightweight CNN architecture that is designed for mobile devices. This study found that MobileNet can achieve high accuracy in apple leaf disease identification using a small model size and low computational cost.

MobileNet has several variants, such as MobilenetV1, and MobileNetV3. Each has improved performance, efficiency, and capabilities. We practiced and examined the MobileNet model using a dataset containing 334 images. Finally, the highest accuracy achieved by this model is 74.

C. Lightweight and Efficient CNN Model

[3] In this study, we suggested a lightweight and efficient CNN model for the identification of apple leaf disease. The model was trained using a collected dataset of apple leaf pictures labeled with different disease types. This study found that the model succeeded in achieving 95.5% accuracy in identifying the apple leaf disease while maintaining a small model size and low computational requirements. It provides a more systematic and practical solution for identifying illnesses of the apple leaf on mobile devices.

To make a lightweight and efficient CNN model, Use fewer layers and fewer filters in each layer. This reduces the complexity of the model. Instead of big filters, use smaller ones. Instead of always reducing the size of the image, find ways to do it only when necessary. This saves computing power. And pick a model that’s designed to be efficient, like MobileNet or SqueezeNet. They are built to be fast and use less memory.

D. Comparison of Plant Leaf Classification Using Modified  AlexNet and Support Vector Machine 

[4 Farmers find it challenging to detect leaf diseases early. As technology advances, many models for detecting diseases have been introduced that are superior to older methods such as apple leaf collection and laboratory process validation. This article uses a pre-trained Model Alexnet using Transfer Learning to classify leaves. A dataset is collected which consists of 3200 images with the measurements of 226 x 226 x 3. Pre-process the collected data and perform the training and evaluation with different variations. Then, modify and apply deep learning techniques like Alexnet which can classify 1000 classes and calculate the performance parameters. The classification can be done based on accuracy and confusion matrix. For this model, the obtained accuracy is 91.15%.

III. MATERIALS AND METHOD

A. Dataset and Pre-processing

The proposed deep Convolution Neural Network model is to predict diseases based on apple leaves, such as apple scab, black rot, cedar rust, and unaffected leaves. A total of 3171 images are collected including healthy and unhealthy leaves of black rot, rust, and powdery apple diseases. A total of 4 labels are created.

The apple leaf images dataset is taken from the Kaggle website which dataset is composed of 3171  images of apple plant leaves.

This model contains four classes of which the three classes correlate with 3 apple diseases namely black rot, scab, and apple cedar rust while the other class contains healthy apple leaf images. The apple leaf images of size 256 × 256 of all four categories were captured with a simple background at various plant development stages. 

It is crucially important to have a wide variety of leaf photos within the collection. Hence, the model can learn important variations throughout the instruction phase. It helpfully assists in enhancing the generalizability of the CNN model. Augmentation presents itself as an approach to artificially create various variations of the image. In this work, amazing transformations like panning, shearing, scaling, zooming, and flipping are used to transform images. These transformations add minor variations in images, which inherently help in introducing variety in the training set. It turns out to assist in reducing overfitting and helps the model achieve better tolerance and an enhanced ability to generalize.

B. Convolution Neural Network

CNN is a feed-forward ANN-based deep learning methodology and is an attractive method. CNN is employed to display more layers within the CNN. Here, we construct a deep CNN by stacking several constructing elements such as convolutional layers, pooling layers, and fully connected levels with typical nonlinear activation units. It has exhibited certain advantages over state-of-the-art ML-based methods like magic! It doesn’t require any additional effort for feature engineering. It is successfully used in several applications including image/text classification, NLP, precision agriculture, etc.

Conclusion

In this work, a CNN model was developed to recognize diseases in apple crops based on apple leaf images. It can assist non-expert farmers in apple orchards and lower the pressure on plant pathologists. Over 1000 epochs were used to train the model on 3171 apple leaves. On the PlantVillage dataset, the model\'s accuracy is evaluated to 98%. The rigorous investigation manifests the proposed model as much better than various pre-trained CNN models. The method was also found superior to other existing models based on various parameters including accuracy and memory requirements. For several diseases, this model achieves good accuracy, ranging from 97% to 99%. The model will successfully balance the accuracy and precision. The AUC-ROC curve shows that the proposed approach is reliable and consistent.

References

[1] Muzafar Rasool Bhat, ‘‘Apple diseases: identification & classification using transfer learning’ 2022.[Online]. Available: https://qascf.com/index.php/qas/article/view/1167/1289. [2] Chongke Bil, “MobileNet Based Apple Leaf Disease Identification” 2020, Available:https://doi.org/10.1007/s11036-020-01640-1 [3] Lili Li;Shujuan Zhang ; Bin Wang; “ Apple Leaf Disease Identification using Lightweight Convolution Networks” 2021, Available:https://doi.org/10.3390/s22010173 [4] Shivali Amit Wagle, Harikrishnan R, “Comparison of Plant Leaf Classification Using Modified AlexNet ” 2021, Available: https://doi.org/10.18280/ts.380108 [5] T. Deshpande, ‘‘State of agriculture in India,’’ PRS Legislative Res., no. 113, pp. 1–29, 2017. [Online]. Available: https://www.prsindia.org/ policy/discussion-papers/state-agriculture-India [6] National Horticulture Board. Report on Apple production in India. Accessed: Dec. 2020. [Online]. Available: https://nhb.gov.in/report_files/apple/APPLE.html. [7] G. Huang, Z. Liu, L. van der Maaten, & K. Q. Weinberger, ‘‘Densely connected convolutional networks,’’ 2016, arXiv: 1608.06993. [8] V. K. Vishnoi, K. Kumar, and B. Kumar, ‘‘Crop disease classification through image processing & machine learning methods using leaf images,’’ in Proc. 1st Int. Conf. Adv. Comput. Future Commun. Technol. (ICACFCT), Dec. 2021, pp. 27–32. [9] A. N. I. Masazhar and M. M. Kamal, ‘‘Digital image processing technique for palm oil leaf disease identification through multiclass SVM classifier,’’ in Proc. IEEE 4thInt. Conf. Smart Instrum., Meas. Appl. (ICSIMA), Nov. 2017, pp. 1–6.

Copyright

Copyright © 2024 B. Ravichandra, P. Shirisha, U. Harshith, Dr. S. Kirubakaran . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.