This project focuses on using machine learning to identify plants, especially medicinal herbs, by analyzing leaf images. It is important because herbal remedies are often cheaper and have fewer side effects than modern medicine. The study reviews techniques for processing leaf images, extracting key features like shape and texture, and using machine learning models to classify the plants accurately. The system tested has shown improved accuracy in identifying plants, which could be useful in real-time applications for herbal medicine identification
Introduction
I. Introduction
Image processing is a technique used to analyze and manipulate visual data (images/videos) using signal processing and mathematical techniques.
The study intersects with computer vision and digital photography, emphasizing the identification of image content using automated systems.
One growing application is in agriculture and botany—specifically, the automatic identification of medicinal plants.
II. Related Work
Prior research has focused on:
Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs) for plant classification.
Using transfer learning and data augmentation to improve performance.
Proposals for real-time detection systems and novel classification models.
Use of leaf vein patterns and high-resolution imagery to enhance accuracy.
Limitations noted include poor handling of complex backgrounds and lack of real-time adaptability in some systems.
III. Proposed Methodology
The system uses deep learning and image processing for automatic detection and classification of medicinal plants via leaf images.
Modules:
Image Acquisition
Focuses on collecting images of herbs from various environments.
Aims to support common users (e.g., farmers, students) without botanical expertise.
Preprocessing
Converts RGB images to grayscale for simplified processing.
Enhances image quality using filters for better feature extraction.
Segmentation
Applies contour models and polygon-based shape definitions to accurately isolate leaves from the background.
Uses contrast mapping to refine leaf edges.
Classification
Utilizes CNN algorithms to identify plant species based on leaf characteristics.
CNNs are chosen for their efficiency in visual pattern recognition and classification tasks.
Usage Details
Once identified, the system provides the user with medicinal uses, plant part references, and availability.
Useful in fields like Ayurveda, pharmaceuticals, and agriculture.
IV. System Architecture
Integrates camera-based image capture, processing pipeline (segmentation, CNN classification), and a user interface to display plant information.
Designed to reduce manual effort and enhance plant identification speed and accuracy.
Conclusion
Traditional medicine methods continue to be widely used in many cases. Population growth, drug shortages, illicit medical costs, side effects of many synthetic drugs, and improved resistance to drugs currently used for infectious diseases has led to increased use of plants as a source of various medicines for human diseases. The project proposed CNN-based methods to obtain Indian leaf varieties. Testing is done with pre -learn and edge detection. CNN is tested with softmax and sigmoid layer. The results confirm that with good edge detection and pre-training, the CNN binary with sigmoid can detect leaf type. The improvement project provides the best and easiest way to separate good trees. The medicinal use of the plant and the great demand for the plant have made it possible to achieve a diversity of plant species.
References
[1] A. Rich, US Keceli, C. Catal, HYYalic, H. Temucin and B. Tekinerdogan, \'\' A transmission analysis study of plant taxa based on a deep neural network, \'\' Comput. Electrons. Farming., Vol. 158, trang 20–29, mars. 2019.
[2] Proceedings of the International Conference on Sustainable Computing and Data Communication Systems (ICSCDS-2022) IEEE Xplore Part Number: CFP22AZ5-ART; ISBN: 978-1-6654-7884-7
[3] J. W. Lee and Y. C. C. Yoon, \'\' identifies a good quality factory using a wide learning model 1, \'\' in Proc. INT. Conf. Technol platform. Services (PlatCon), January 2019, p. 1 to 5.
[4] M. Arsenovic, M. Karanovic, SSladojevic, A. Anderla and D. Stefanovic, “Addressing current limitations of deep learning methods based on plant disease diagnosis,” Symmetry, vol. 11, no. 7, p. 939, juillet 2019.
[5] H. S. Abdullahi, R. ESheriff and F. Mahieddine, \"Legal Neural Networks in Precision Agriculture for Image Recognition and Crop Classification\", in Proc. Int. 7 Confessions. Creation. Computer. Technology. (INTECH), August 2017.
[6] V. Bodhwani, D. P. Acharjya and U. Alone, ‘‘Remedies for Deep Crop Identification Network’’ Proc. Computer. Science Fiction, Vol. 152, pp. 186–194, Jan. 2019
[7] I. Z. Mukti et D. Biswas, ‘‘Learning Transfer Based on Plant Diagnostics Using ResNet50,’’ at Proc. Int. 4 Confession. Vote. Inf. Community. Technology. (EICT), December 2019, pp. 1–6 [8]L. C. De, “Biodiversity and conservation of aromatic plants” Adv. Agricultural crops. Res., Vol. 5, no. 4, p. 00186, December 2016[2] MM 235, pp. 228–235, medium. In 2017.