A Comprehensive Survey on Weed Identification in Agriculture using Machine Learning

Abstract

Unchecked weed growth can seriously affect crop yield and quality. Excessive use of herbicides to control weed growth is harmful to the environment. Identifying areas infested with weeds helps in the selective chemical treatment of those areas. Similarly, we can also implement precision spraying techniques for the crops. Advances in farm image analysis have created a solution for identifying weedy plants. However, these are supervised learning methods that require many manually annotated images. Hence, these approaches are not economically feasible for individual farmers due to the wide variety of crop species grown. In this review, algorithms, such as CNN and CNN-based algorithms, K-Means, SVM, Fuzzy algorithms, Hough transform and Gabor filter and others to accurately estimate weed distribution and density are covered in detail. Deep-learning-based methods to robustly estimate weed density and distribution are discussed in detail in this review. In this paper, an overview of image segmentation methods, detection approaches and various classification techniques are identified. Further, the existing solutions are presented with their own challenges.