Weed and Crop Classification with Domain Adaptation for Precision Agriculture

Abstract

Agricultural chemicals, which are frequently used in agricultural fields, have many different side effects on the environment, biological diversity and people. It is aimed to reduce the use of agricultural chemicals within the scope of sustainable agriculture. For this purpose, it is an important requirement for robots used in agricultural fields to precisely distinguish between crops and weeds. One of the biggest challenges in this regard is the wide variety of plant species. For this reason, the systems to be used in agricultural areas need to cope well with the changes caused by weed and crop varieties. In this study, a examination is made using various unsupervised domain adaptation approaches to tackle the challenge of species diversity in the weed and crop classification task. Thus, it is aimed to measure the generalize the classification process of crop and weed species that have not been seen before.