Sauvola Segmentation and Support Vector Machine-Salp Swarm Algorithm Approach for Identifying Nutrient Deficiencies in Citrus Reticulata Leaves

Lia, Kamelia (2024) Sauvola Segmentation and Support Vector Machine-Salp Swarm Algorithm Approach for Identifying Nutrient Deficiencies in Citrus Reticulata Leaves. Doctoral thesis, Asia e University.

	Text Thesis Lia Kamelia.pdf Restricted to Repository staff only Download (5MB)
	Text Thesis Lia Kamelia-1-24.pdf Download (392kB)

Abstract

Machine learning and image processing methods can effectively detect nutrient deficiencies in citrus trees, addressing the challenge of accurately identifying shortages that can impair crop health and productivity. Traditional methods often rely on expert visual assessments, which are labour-intensive, subjective, and time-consuming. The proposed method integrates colour and texture feature-based image analysis with machine learning algorithms for classification. The process begins with acquiring image data, which is categorized into four classes: nitrogen (N) deficiency, phosphorus (P) deficiency, potassium (K) deficiency, and normal. In total, 1,200 images are collected. Next, file sizes are reduced using lossless compression methods, achieving a 96.99% reduction. The second phase involves image segmentation using the Sauvola method. Following this, colour and texture featureextraction is performed. Colour features are extracted in the Hue (H), Saturation (S), and Value (V) colour space, while texture features are obtained using the Grey-Level Co-Occurrence Matrix (GLCM) method. This combination of colour and texture features results in various metrics, including mean, dissimilarity, skewness, angular second moment, variance, entropy, maximum probability, contrast, correlation, energy, and homogeneity, which are used for classification. Both Support Vector Machine (SVM) and Artificial Neural Network (ANN) methods are compared for classification. The Sauvola method combined with ANN achieves the highest accuracy of 93.75%. In the next phase, the datasets are optimized using the Salp Swarm Algorithm (SSA), which improves classification accuracy. With SSA optimization, the Sauvola method combined with SVM reaches an accuracy of 99.58%, surpassing other methods that use image processing and ANN classification. Expert validation is utilized to evaluate and validate the effectiveness of the proposed method and confirm the system's accuracy at 95%. Integrating SSA and SVM machine learning algorithms improves decision-making processes, leading to better crop yield through early detection and timely nutrient management. It ensures that plants receive the necessary nutrients for optimal growth and development.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Citrus leaves, classification, Sauvola segmentation, optimization, Salp Swarm Algorithm
Divisions:	School of Graduate Studies
Depositing User:	Siti Nor Fairuz Rosaidee
Date Deposited:	27 Dec 2024 07:44
Last Modified:	27 Dec 2024 07:44
URI:	http://ur.aeu.edu.my/id/eprint/1248

Actions (login required)

View Item