Weed Identification in Crops Based on Convolutional Neural Networks
Download as PDF
DOI: 10.23977/iset.2019.049
Corresponding Author
Shaobo Liang
ABSTRACT
This study aims to develop an algorithm of transfer learning based on the Convolutional Neural Network Inceptionv3 for weed recognition by using images of weeds. Inceptionv3 shows great aptitude of image sorting as the basis of the network; classifiers are added onto Inveptionv3, weight initialization of transfer learning is implemented, and the model is trained through 1000 iterations with images downloaded from the open sources data on Kaggle. After the training, the model performed phenomenally, with a over 99% accuracy in the training set and a almost 90% accuracy in the validation set. The model also shows that it is well functioning in the actual validation, correctly recognizing the species of the weeds. This paper provides a possible solution to the need of weed recognition in precision agriculture.
KEYWORDS
Weed recognition, Weed image, model