Recognition of weeds in corn crops: System with convolutional neural networks

Abstract

This work proposes weed classification solutions in corn crops using Deep Artificial Neural Networks. The InceptionV3, MobileNetV2 and Adapted MobileNetV2 Convolutional Neural Network architectures were used to extract features from the images. The research problem refers to the question “Does the use of Artificial Neural Networks present good performance for visual analysis in the recognition of weeds in corn crops?”. In this way, the research was based on the analysys of image recognition methods and techniques with Artificial Neural Networks and organizing a database with images of corn crops, training and comparing the accuracy results and analyzing the behavior of different architectures on the dataset. Among the models, the one with the highest accuracy was InceptionV3, with 98%, demonstrating the ability of the technologies used in this work to become realistic for applications in crops, in order to help farmers to increase productivity in their crops.