Detection of Sigatoka Disease in Plantain Using IoT and Machine Learning Techniques

Abstract

Achieving United Nations Sustainable Development Goal 2 (UN SDG2) infers an imperative to urgently increase food production by up to 70%. However, concerns have risen that increases in food production have not kept pace with increase in world population, which is estimated to reach 10 billion people by the year 2050. In this paper, an IoT with machine learning based system was developed to acquire and process significant indicators such as temperature, moisture, humidity and leave images for the detection of Sigatoka disease in plantain. Appropriate sensors for detecting the stated disease indicators were interfaced with Raspberry Pi3 microcontroller module to collate and transmit the sensor data wirelessly to ThingSpeak, which is the selected cloud based IoT platform. The acquired leave images were further processed using two image descriptors, namely: Scalable Color Descriptor (SCD) and Histogram of Oriented Gradient (HOG) to extract discriminative color and texture features respectively. The features were then classified to detect the diseased or non-diseased class using Multilayer PerceptronArtificial Neural Network (MLP-ANN). The best accuracy of 98% was produced using the HOG descriptor.