Malaysia's palm oil industry powers the economy, generating RM109 billion in exports during 2024 and contributing around 3% to national GDP. Smallholder farmers overall have 1.48 million hectares of plantations which over 450,000 individuals. Rural areas encounter obstacles in swift disease identification and crop safeguarding, stemming from delayed approaches, elevated expenses for advanced tools and specialists, and dependence on internet connectivity in isolated locations. Ganoderma Basal Stem Rot (BSR) inflicts yearly economic damages up to RM2.2 billion. An AI-driven framework detects system in palm oil trees and leaf, categorizing 12 primary states encompassing mild, moderate, and severe Ganoderma Basal Stem Rot, foliar infections from Cephaleuros, Curvularia, and Drechslera, shortages of boron and potassium, plus sound stems and leaves, delivering 99.88% accuracy at inference speeds below 1 second. Operating without internet, cost under RM600, portable and high accurate, early alerts elevate output and eco-friendliness. Alignment occurs with SDG 8 promoting economic expansion and rural employment, SDG 9 closing technological divides in distant estates, and SDG 15 fostering robust plantations to curb epidemic outbreaks and forest depletion.

Invention Technical Description

The overall proposed block diagram for the computational procedure of plant disease detection. The model leverages a pre-trained MobileNetV3Large-based CNN model, chosen for its high speed and accuracy, as demonstrated by research comparisons. The pre-trained layers (base model) were frozen without modification to extract features. The CNN architecture incorporates numerous layers for feature extraction with the pre-trained MobileNetV3Large. Two dense layers, with 128 neurons and 64 neurons respectively, were applied, which sum the weighted inputs and apply the ReLU activation function. This non-linearity allows the model to learn complex relationships between features and disease classes. To prevent overfitting, a Dropout layer was added for each dense layer, with 50% of neuron outputs being randomly deactivated. This forces the network to learn more robust features during the training process, thereby reducing the risk of the model relying too heavily on any single neuron. Finally, a fully connected layer was used as the classification output, identifying six distinct palm oil disease classes

Demostration/ Presentation Video

Poster/ Broucher/ Invention Photo