Detection and classification of insects on stick-traps in a tomato crop using Faster R-CNN
Nieuwenhuizen, A.T.; Hemming, J.; Suh, H.K.
In this paper we present the method and performance to detect tomato whitefly and its predatory bugs on yellow sticky traps. These traps are imaged in controlled light conditions with a digital single lens reflex camera and in uncontrolled environment with smartphone camera. The method consists of the following steps. First, image sub setting and data labelling by manual annotation. Secondly, training a deep learning convolutional neural network. Third step is classification of the images. Final step is comparison with hand counted data of insects. The weighted averaged accuracy for deep learning detected insects was 87.4%. The correlation of hand counted insects with deep learning counted insects was over 0.95 for the smartphone images. The methods used show that the training data used on controlledconditions could be transferred to uncontrolled smartphone imaging conditions for the data provide