Plant Leaf Disease Classificiation using CNN new.pptx

This presentation is about Plant disease classification.

1. Deep Learning Based Plant Disease Classification using Plant Village dataset By: Bhanu Pratap Singh

2. Outline Introduction Materials and Methods Results Limitations and Future Scope Conclusion References

3. Introduction Crop production worldwide is adversely affected by various plant diseases, highlighting the importance of disease identification and classification. Traditionally, this process relied on specialists and experts, which proved to be time-consuming and cumbersome. Convolutional Neural Networks (CNNs), particularly in the domain of plant leaf diseases, have demonstrated remarkable success. This study focuses on training pre-trained models using various plant leaf diseases. Additionally, the study explores the impact of data augmentation, fine-tuning, and dropout techniques on model performance.

4. Research objective Work on popular plant diseases dataset known as Plant Village. U se various CNN models to classify them. Transfer Learning i.e. p re-trained models will be used. Further the data augmentation techniques, fine-tuning and dropout will be carried too. Then the models can be compared based on various metrics like accuracy, precision, recall, f1 score.

5. System Requirements and Framework The proposed research needs good computational device with a powerful GPU. Kaggle Notebooks were used for the training of the networks. Kaggle provides free GPU sessions for about 30 hours/week. The GPU P100 was used. Python and Tensorflow framework for this research TensorFlow is a free and open-source software library for ML and AI. It can be used across a range of tasks but has a particular focus on training and inference of deep neural networks. Materials and Methods

6. Dataset Dataset Number of Classes Total Images Description Plant Village [1] 39 55,448 The dataset contains many plant leaf diseases and is the most famous dataset for the classification of plant diseases.

7. Model Year Parameters (Approx.) Advantages Disadvantages MobileNetV2 [2] 2018 3.4 million Lightweight architecture for mobile and embedded devices, Efficient use of depthwise separable convolutions, Good balance between accuracy and model size May sacrifice some accuracy compared to larger and deeper models InceptionV3 [3] 2015 23.8 million Utilizes inception modules for multi-scale feature extraction, Strong performance on image classification tasks, Suitable for medium to large datasets A relatively larger number of parameters compared to some newer architectures Xception [4] 2016 22.9 million Utilizes depthwise separable convolutions for efficient feature extraction, High accuracy on various image classification benchmarks, Increased model depth may lead to higher memory and computational requirements DenseNet201 [5] 2017 20.2 million Densely connected layers promote feature reuse and gradient flow, Strong accuracy and parameter efficiency, Suitable for medical image analysis and other tasks May require more memory and computational resources compared to shallower architectures CNN Models

8. Limitations and Future Scope Limitations: Test set derived from the dataset itself, which may not fully represent real-world scenarios. Images are taken in very good conditions with no noise but results might differ when real time images are used. Limited coverage of diseases and plants. Future Scope: Explore advanced machine learning and deep learning methods. Optimize models for real-time implementation on edge devices and smartphones, enhancing usability for farmers. Collaborate with agricultural specialists, plant pathologists, and machine learning researchers to develop specialized and efficient solutions. Multidisciplinary cooperation can uncover disease-specific indicators and patterns not visible through data-driven approaches alone.

9. References [ 1 ] Hughes, David, and Marcel Salathé . "An open access repository of images on plant health to enable the development of mobile disease diagnostics." arXiv preprint arXiv:1511.08060 (2015). [2] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov , and L. C. Chen, "Mobilenetv2: Inverted residuals and linear bottlenecks," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 4510-4520. [3] C. Szegedy , V. Vanhoucke , S. Ioffe , J. Shlens , and Z. Wojna , "Rethinking the inception architecture for computer vision," in Proceedings of the IEEE Conference on computer vision and pattern recognition, 2016, pp. 2818-2826.G. Eason, B. Noble, and I. N. Sneddon, “On certain integrals of Lipschitz-Hankel type involving products of Bessel functions,” Phil. Trans. Roy. Soc. London, vol. A247, pp. 529–551, April 1955. (references) [4] F. Chollet, "Xception: Deep learning with depthwise separable convolutions," in Proceedings of the IEEE Conference on computer vision and pattern recognition, 2017, pp. 1251-1258. [5] G. Huang, Z. Liu, L. Van Der Maaten , and K. Q. Weinberger, "Densely connected convolutional networks," in Proceedings of the IEEE Conference on computer vision and pattern recognition, 2017, pp. 4700-4708.

10. Thank You