Effectiveness of convolutional layers in pre-trained models for classifying common weeds in groundnut and corn crops

Abstract

In modern agriculture, herbicides are most commonly used to control weeds. A large amount of herbicide usage not only has an adverse effect on the soil but also has a severe environmental impact. A smart herbicide sprayer combines the features of a selective and variable rate sprayer to control the amount of herbicide used. Its efficiency primarily depends on accurately identifying the weed type and density. The main objective of this study was to detect weeds using various deep learning techniques. This study provides a new balanced and multi-class dataset for groundnut crops with 15 frequent weeds taken under various lighting conditions and at different places with less noise. There are 24,816 images in the groundnut weed dataset (GWD), and each image contains a single plant. Finally, we compared the performance with and without freezing the convolutional layers in five pre-trained architectures: AlexNet, VGG-16, VGG-19, ResNet-50, and ResNet-101 on the new balanced groundnut weed dataset (16-classes) and the existing balanced corn weed dataset (5-classes). However, ResNet-50 and ResNet-101 on the groundnut and corn weed datasets produced the highest accuracies of 99.84% and 100%, respectively. VGG-19 is an extensive network that learns more features without freezing convolutional layers than the remaining pre-trained models.