Elimination of Herbicides after the Classification of Weeds Using Deep Learning

Abstract

Introduction: Herbicides are chemicals that are used to destroy weeds. It is commonly used in agriculture to kill undesired plants and increase crop yield, even though it has negative effects on humans and the environment. Pesticides sprayed on crops must be decreased in the real world to protect humans, animals, and birds from dangerous diseases such as cancer, eyes, and skin infection. Pesticides are classified as herbicides. Deep learning is being used in this research to minimize chemical compounds. Scientists seek to limit the amount of pesticide sprayed on crops to protect humans and the environment from toxic exposure. Background: In this research, A neural network classifier is built using Convolutional Neural Network (CNN), dropout, rectified linear activation unit (ReLU), the Root Mean Squared Propagation (RMSprop) optimization technique, and stochastic gradient descent (SGD). The algorithms based on CNN outperformed the others. This study uses generated dataset (unique dataset and processes it rowwise through the Neural network) to train a categorized neural network, and the dataset was created with the assistance of the agriculture professor. Methods: This study offers a method for classifying weed images and spraying herbicides solely on weeds/unwanted plants rather than crops. The model should first be trained using the training dataset before being tested using the testing datasets. Results: This model's training accuracy is 96%, while testing accuracy is 89%. Conclusion: This model reduced herbicide (it is a type of pesticide/chemical) spray over the crop (foods, vegetables, sugarcane) to protect humans, animals, birds, and the environment from harmful chemicals.