Publication: Machine learning-based image processing and iot-based monitoring system for disease detection in harumanis mango plantations.
No Thumbnail Available
Date
2025-11
Authors
Sheikh Mohamad Naim bin Shikh Shatir
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Local farmers face multiple challenges in managing Harumanis plantations. The effect of Movement Control Order (MCO) has restricted farmers from closely
supervising their crop plantation, while, the lack of utilization of smart farming (SF) devices among farmers in Malaysia is due to the high cost of such devices.
Consequently, many farmers persist with evaluating soil nutrient and disease recognition using traditional method, while new farmers in particular, struggle to
differentiate crop diseases due to limited agricultural knowledge. The goals of this research are to develop a cost-efficient, IoT-based monitoring system with image processing algorithm, evaluate parallel image processing techniques, decide the most suitable algorithm for leaf disease detection supported by analyzed results and correlate extracted feature patterns for corresponding leaf diseases with nutrient parameters—temperature, moisture, and nitrogen, phosphorus and potassium (NPK) values, obtained from sensors. The sensors are connected to an ESP8266 microcontroller, in which the data are transmitted through wireless communication to a Raspberry Pi (RPi) microprocessor. A custom camera attached to the RPi will provide image of the Harumanis mango plant, where the images are processed using Open Computer Vision (OpenCV) library for detection and classification of nutrient deficient disease affecting the Harumanis mango plant. Additionally, the Normalized Difference Vegetation Index (NDVI) value is analyzed. Furthermore, a graphical user interface for Internet of Things (IoT) system is built and accessible from end-devices, for crop monitoring. The IoT system is integrated with parallel processing for Harumanis mango leaf diseases detection, using YOLOv8 model, trained on custom dataset. The system incorporates machine learning object-detection algorithm, capable in extracting harumanis leaf diseases features— anthracnose, leaf burn, and healthy leaves. The ML performance trained with augmented dataset improved the detection
by 14% for anthracnose, 5% for healthy and 3% for leaf burn. Using logistic regression (LR), the system performed correlation analysis based on leaf disease count and sensor parameters (temperature, humidity, soil moisture and NPK). The model then predicted that a healthy Harumanis mango plant, under experimental conditions of this study, have nitrogen value of 30 mg/kg, phosphorus is 47 mg/kg and potassium is 125 mg/kg. These values were based on the correlation analysis between extracted Harumanis mango leaf disease counts and sensor measurements collected during field trials, ensuring that the nutrient thresholds are derived directly in this research dataset than assumed from literature.