Crop Health Detection Using Scale-Invariant Feature Transform (SIFT) and Oriented FAST and Rotated BRIEF (ORB)

Abstract

Agriculture plays a vital role in every nation, as a healthy population relies on robust yields to ensure food security. With the continuous growth of the population, the excessive use of pesticides and fertilizers has become prevalent, which can negatively impact crop health. In this paper, we introduce a solution that utilizes images of crops, employing image processing and machine learning techniques to classify them as healthy or unhealthy. Various feature detection and extraction methods are available, but we specifically compare Oriented FAST and Rotated BRIEF and scale-invariant feature transform in this work. Both techniques can effectively extract features from the images, and we can use the matcher function from OpenCV to determine if the extracted features correspond to those of a trained image. If there is a match, it indicates an unhealthy crop, while a lack of matching suggests the crop is healthy. Additionally, machine learning classifiers can be employed to enhance training on these extracted features, leading to improved results and predictions.

Introduction

Healthy agriculture is crucial for a country’s economy and food security, but crop diseases pose significant challenges by reducing yield and quality. Traditional manual detection methods are labor-intensive and error-prone, highlighting the need for automated, efficient disease detection systems. Recent advances in image processing and machine learning offer promising solutions. Techniques such as SIFT (Scale-Invariant Feature Transform) and ORB (Oriented FAST and Rotated BRIEF) are effective for feature extraction from crop images, enabling classification of crops as healthy or unhealthy. SIFT provides high accuracy and robustness to scale, rotation, lighting, and noise variations, while ORB is faster and computationally efficient, making it suitable for real-time applications.

The research methodology involves:

1. Data Collection: Capturing high-quality images of healthy and diseased crops under varied conditions.

2. Image Pre-processing: Resizing, grayscale conversion, and noise reduction.

3. Feature Extraction: Using SIFT and ORB to detect keypoints and descriptors for disease-related patterns.

4. Feature Matching & Classification: Comparing extracted features using KNN or FLANN to classify crops as healthy or diseased.

5. Machine Learning Integration: Training classifiers (SVM, random forest, CNNs) on extracted features for end-to-end disease prediction.

Experiments on wheat images showed that both SIFT and ORB successfully detected disease features, with ORB outperforming SIFT in speed and number of features extracted. While SIFT is highly accurate and robust, it is slower and has computational limitations for real-time applications.

Conclusion

ORB is a more computationally efficient alternative to SIFT, designed to offer similar performance with a lower computational cost. It combines the FAST keypoint detector and the BRIEF descriptor, both of which are faster and more efficient than their counterparts in SIFT. However, ORB may not be as precise as SIFT in cases where the disease patterns are very subtle or require very high-detail extraction. SIFT is preferable when accuracy and robustness to variations in crop images are paramount, especially when dealing with diverse environmental conditions. ORB is more suitable for real-time applications and environments where computational efficiency is critical, while still providing acceptable accuracy for crop disease classification. A hybrid approach that combines both feature extraction techniques with deep learning models could also be explored to leverage the strengths of both methods. This could potentially lead to even more robust and accurate crop disease detection systems, capable of scaling to large agricultural datasets. Different methods with hybrid approach can enhance the results as SIFT, ORB, CNN, and deep learning. Different environmental conditions (lightning, noises in image, angles of capturing, etc.) of acquiring dataset can also be considered in future to enhance the application.

References

[1] Prashanthi T, Susmitha T, Mishra RK: Crop disease detection using NLP and deep learning. Proceedings of the 2nd International Conference on Cognitive and Intelligent Computing. Kumar A, Ghinea G, Merugu S (ed): Springer, Singapore; 2023. 197-203. 10.1007/978-981-99-2742-5_21. [2] Bhagwat R, Dandawate Y: Towards robust crop disease detection for complex real field background images. Vietnam Journal of Science and Technology. 2024, 62:990-1004. 10.15625/2525-2518/18914 [3] Paithankar V, Awari A, Raskar A, Patil S, Jamdar N: Plant disease detection using machine learning. International Journal of Advanced Research in Science Communication and Technology. 2023, 3:267-272. 10.48175/ijarsct-9297. [4] Rabbi MMH, Hossain MJ, Masud MA, Rahman MA: An image recognition approach for crop disease detection in agro-field from infected plant area. Global Mainstream Journal of Innovation, Engineering & Emerging Technology. 2024, 3:11-18. 10.62304/jieet.v3i05.219 [5] Batham S, Sharma AK, Rathore AS: A research paper on crop disease detection using deep learning model. International Journal for Research in Applied Science & Engineering Technology. 2023, 11:399-406. 10.22214/ijraset.2023.56002 [6] Neha Sharma, Pradeep Kumar: Crop Health Detection Using Image Processing and Machine Learning for Better Yield Production, published on Cureus Journal of Computer Science

Copyright

Copyright © 2025 Sayali Joshi, Dr. Mohammad Atique M. Junaid. 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.