Classification Model Evaluation with Feature Extraction for Pest Attacks Identification on Cocoa Pods

Abstract

This study utilizes Artificial Neural Network (ANN) models for cocoa pod pest attack detection, emphasizing tailored models for image feature matching. Two classification models using Local Binary Pattern (LBP) feature extraction are evaluated. Computer vision analyzes fruit characteristics in images, distinguishing between normal and pest-infested conditions. The study also considers computational efficiency for real-time drone-based identification. It involves 220 images for both categories, using LBP for feature extraction, and evaluates Multilayer Perceptron (MLP) and Extreme Learning Machine (ELM) approaches. ELM outperforms with 66% accuracy in just 0.592543 seconds, notably with the Softlim Activation Function and 100 hidden layers. Z-Score analysis supports ELM's superiority, especially with the Sine Activation Function and 50 network nodes, offering valuable insights for cocoa pest detection and highlighting the effectiveness of ELM as an ANN model.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint