The Identification of Weeds and Crops Using the Popular Convolutional Neural Networks

Abstract

The weeds compete with the crops for the nutrients. However, it is necessary to control weeds in order to prevent them from affecting the productivity of the land. In the context of the robotization of this process, the Deep Learning can help to localize and identify the weeds, thanks to the object detector. The accuracy of an object detector depends on its backbones which are networks their function is feature extraction. In this research, we have shown the performance of the most popular CNNs, which are used as backbones, for the identification and distinction of crop weeds, as well as proposed a model that combines the architecture of the two models MobileNet and ResNet. The proposed model had the properties of MobileNet in terms speed and precision, our modification has resulted in a significant improvement. The obtained results, showed us that the MobileNet models are the most suitable for as backbone to the future object detector that we want to design it to detect and ultra-localize the weeds.