Digital image processing for preliminary detection of infected porang (Amorphophallus muelleri) seedlings

Abstract

Porang (Amorphophallus muelleri) is an Indonesian parental plant tuber developed vegetatively from bulbils during dormancy and harvested through petiole detachment for the industrial production of glucomannan. Pathogenic fungi and whiteflies can cause infection during harvesting and storage, destructing plant cells as well as reducing seed quality and crop yields. Therefore, this study aimed to develop a calibration model for detecting infected and non-infected porang bulbils using a computer vision system. Image parameters such as colour (red, green, blue - RGB and hue, saturation, intensity - HSI), texture (contrast, homogeneity, correlation, energy, and entropy), and dimensions (width, area, and height) were evaluated on 90 samples in three positions. The results showed that the majority of image quality properties were significantly associated with non-infected and infected porang bulbils as showed by Pearson correlation values of 0.901 and 0.943, respectively. Discriminant analysis based on image attributes effectively classified non-infected and infected seedlings, achieving a model accuracy of 97.0% for correctly classified cross-validated grouped cases. Therefore, computer vision can be used for the preliminary detection of fungal infection in porang bulbils, as evidenced by its high accuracy and outstanding model performance.

Keywords

• discriminant analysis
• gray-level cooccurrence matrix
• model performance
• seed quality
• vegetative phase