An Evaluation of Machine Learning Techniques for Crop Detection from Garden Images

Abstract

Agriculture is a major driver for different economies across the globe. With the continued advancement in technology, there is a move towards precision agriculture. A major area of research interest is the detection of crops by use of computer vision. Such can help in controlling pests and diseases and thus increase yields. However, given the wide range of techniques applied, there is a need to evaluate them to recommend the most suitable for the detection of different crops. This study was thus aimed at evaluating both traditional Machine Learning (ML) and Deep Learning (DL) techniques for the detection of banana and cassava crops from a set of garden images. First, a crop classification model was built using the traditional machine learning model with feature extraction algorithms being applied before training using K-Nearest Neighbor (KNN) and Naïve Bayes. Another crop classification model was also built using deep learning (CNN) on an annotated dataset. The performance of both models was evaluated with deep learning giving the best result. Deep learning is thus recommended as the best model for crop detection.