Citrus Tree Nutrient Deficiency Classification: A Comparative Study of ANN and SVM Using Colour-Texture Features in Leaf Images

Abstract

Nutrient deficiency in Citrus reticulata (mandarin orange) plants causes reduced plant resistance against diseases and pests. This research presents a combined approach to identify nutrient deficiencies in Citrus Reticulata var. Fremont leaves using image processing and machine learning techniques. This study uses leaf images to accurately and efficiently detect nutrient deficiencies in mandarin orange plants. The image data is divided into four classes: normal, N-minus, P-minus, and K-minus. The file sizes are compressed using a lossless compression method, resulting in an average file size reduction of 96.99%. Subsequently, the images undergo contrast stretching to improve their quality. Parameters such as Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) are measured. The maximum PSNR is 35.801412680386456, and the minimum PSNR is 14.011790825259139, with a good range of 25-30 dB for PSNR. The SSIM scores after compression and contrast stretching are 0.9734938845160109 (maximum) and 0.8860099106663607 (minimum), which fall within the good range of 0.8-0.9. The second stage involves applying segmentation processes to the images using the Canny and Sauvola methods. Canny effectively identifies sharp and clear edges, while Sauvola retains image details, making it more suitable for texture and colour feature extraction. The third stage involves extracting colour and texture features from the images. Colour feature extraction is done using the H (Hue), S (Saturation), and V (Value) colour space. Texture feature extraction utilises the Grey-Level Co-Occurrence Matrix (GLCM) method. The feature values will be used for the classification process in the next stage. The fourth stage involves the classification process based on the segmentation results using the Canny and Sauvola methods, performed separately using Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods. These process results in four datasets: Canny-ANN, Canny-SVM, Sauvola-ANN, and Sauvola-SVM. The highest accuracy is achieved by the Sauvola-ANN method, with a value of 93.75%.