Weed scouting is an important part of modern integrated weed management butcan be time consuming and sparse when performed manually. Automated weedscouting and weed destruction has typically been performed using classificationsystems able to classify a set group of species known a priori. This greatlylimits deployability as classification systems must be retrained for any fieldwith a different set of weed species present within them. In order to overcomethis limitation, this paper works towards developing a clustering approach toweed scouting which can be utilized in any field without the need for priorspecies knowledge. We demonstrate our system using challenging data collectedin the field from an agricultural robotics platform. We show that considerableimprovements can be made by (i) learning low-dimensional (bottleneck) featuresusing a deep convolutional neural network to represent plants in general and(ii) tying views of the same area (plant) together. Deploying this algorithm onin-field data collected by AgBotII, we are able to successfully cluster cottonplants from grasses without prior knowledge or training for the specific plantsin the field.
展开▼
