Deep Weed Detector/Classifier Network for Precision Agriculture

Abstract

The productivity of crop farming keeps diminishing at an alarming rate due to infestation of weeds and pests. Deep learning is becoming as the approach for identifying weeds on farmlands. However, training weed data sets with deep learning classification alone trains the whole images consisting of the weed and its background (soil) without categorically telling which particular item in the image is a weed. This makes utilising this classification approach for precision agriculture difficult. We present an alternative approach, which involves incorporating a pre-trained network in this case ResNet-50 and YOLO v2 object detector for weed detection/classification on farmlands. Thus, weeds can precisely be located, identified (type), sprayed with the appropriate herbicide or removed with the appropriate mechanism. This sums up weeding process in precision agriculture.