Research Article
Weed Identification in Plant Seedlings Using Convolutional Neural Networks
@INPROCEEDINGS{10.1007/978-3-031-35883-8_14, author={Samuel Damilare and Chika Yinka-Banjo and Olasupo Ajayi}, title={Weed Identification in Plant Seedlings Using Convolutional Neural Networks}, proceedings={Emerging Technologies for Developing Countries. 5th EAI International Conference, AFRICATEK 2022, Bloemfontein, South Africa, December 5-7, 2022, Proceedings}, proceedings_a={AFRICATEK}, year={2023}, month={7}, keywords={Precision Agriculture (PA) Deep Learning Convolutional Neural Network (CNN) Weed Control Plant Seedling Classification Transfer Learning}, doi={10.1007/978-3-031-35883-8_14} }- Samuel Damilare
Chika Yinka-Banjo
Olasupo Ajayi
Year: 2023
Weed Identification in Plant Seedlings Using Convolutional Neural Networks
AFRICATEK
Springer
DOI: 10.1007/978-3-031-35883-8_14
Abstract
Agriculture is essential for the continuous survival of man, however, the adverse effect of weeds in agronomy cannot be ignored. These weeds compete with crops for nutrients and sunlight, hence resulting in low crop yield. It is therefore necessary to identify and remove them at an early growth stage for effective weed control and maximum farm produce. This study focuses on distinguishing between crops and weeds at their infancy, using images processing. To achieve this, three convolutional neural networks (CNNs) architectures, ResNet, MobileNet and InceptionV3, were evaluated using a transfer learning technique on a dataset of 5,339 RGB plant images containing 12 different species of plants. Comparing their performances from experiments carried out, the results revealed the Inception V3 model as the best for crop identification with an accuracy of 82.4%, while ResNet and Mobilenet both achieved average accuracies of 71.1% and 75.4% respectively. ResNet however gave the best performance in terms of identifying weeds. Overall, Inception v3 was the best as other performance metrics including the recall, precision, and F1-score also corroborated the superiority of Inception v3 in distinguishing between crops and weed.
