Through this journey, a comparative study was conducted on ResNet-50, ResNet-34, and Inception-v3 architectures to solve the problem of distinguishing between eight weed species from over 17,000 images.
The selected weed species are local to pastoral grasslands in Queensland and include Chinee apple, Snakeweed, Lantana, Prickly acacia, Siam weed, Parthenium, Rubber vine, and Parkinsonia.
Pretrained Inception-v3 achieved the highest validation accuracy (92%) and the lowest validation loss (~0.2), indicating superior generalization.
A detailed classification report revealed key findings about the model performance.