Investigating the Efficacy of Feature Extraction-Neural Networks and Attention Based Deep Learning Models for Crop Weed Detection

Abstract

The domain of agriculture is witnessing a drastic change in terms of technological use and precision agriculture which aims at leveraging data analytics for agricultural applications. One critically important application of precision agriculture happens to be crop-weed detection based on machine learning and deep learning-based algorithms. One of the major challenges which automated crop-weed detection algorithms face is the effects of noise and disturbances typically through images captured through unmanned aerial vehicles (UAVs). This paper presents an analysis of feature extraction followed by machine learning for detecting crop weeds. Noise removal and feature extraction has also been employed to bolster the training process. Alternatively, an attention based deep learning model has also been developed for weed detection. The attention-based approach has been developed with the aim of identifying the most critical information from large datasets which has the potential to enhance the training efficiency of the approach. A comparative analysis of both approaches in terms of classification accuracy has been presented. The experimental results clearly show that the proposed approach outperforms contemporary deep learning algorithms such as CNN, ResNet, YOLO and RCNN in terms of classification accuracy.