Weed/Plant Classification Using Evolutionary Optimised Ensemble Based On Local Binary Patterns

Abstract

This thesis presents a novel pixel-level weed classification through rotation-invariant uniform local binary pattern (LBP) features for precision weed control. Based on two-level optimisation structure; First, Genetic Algorithm (GA) optimisation to select the best rotation-invariant uniform LBP configurations; Second, Covariance Matrix Adaptation Evolution Strategy (CMA-ES) in the Neural Network (NN) ensemble to select the best combinations of voting weights of the predicted outcome for each classifier. The model obtained 87.9% accuracy in CWFID public benchmark.