BOHR International Journal of Smart Computing and Information Technology

The purpose of this study is to compare the accuracy of several deep-learning models for the identification of
rice weed. In this study, 1500 datasets of local rice and 1000 datasets of weed were resized and applied to the
input size of the network, respectively. A total of 70% of the data were used for training, and the remaining 30%
were used for validation. MATLAB R2018a was used to construct the AlexNet pre-trained model using a transfer
learning strategy, and by changing the AlexNet model, RiceWeedNet, a convolutional neural network, was created.
Metrics such as network accuracy, recognition accuracy, precision, and recall were used to assess both models’
performances. While the test set’s identification accuracy is 97.713415%, its precision is 0.9776, and its recall
value is 0.9803. The RiceWeedNet model achieved a network accuracy of 100%. A network accuracy of 90% and
a recognition accuracy of 73.780488% were reported by the AlexNet model, respectively. The created model may
be used instead of conventional weed detectors.