Sticky Traps with insects for training of deep learning Convolutional Neural Networks for object detection.
How to use?
On yellow sticky traps that were hanging in commercial greenhouses, insects were collected. The following insects were annotated on the yellow sticky traps: Whitefly, Macrolophus and Nesidiocoris. The dataset is suited to training deep learning convolutional neural networks for object detection.
Insects detection and classification on stick-traps in tomato crops is of relevance for proper pest management and reduced use of inputs. Counting and classification of insects is however time consuming and error prone. Typically, the counting of insects on the traps is done manually. More recently, methods have become available to partly automate the counting. Specific hardware records the so called yellow sticky traps and counts and identifies insects. However more accurate numbers of insects are required for population models and other cameras like smart phone cameras would increase the uptake of this technology by farmers more easily. Therefore the idea was launched that trained operators provide labelled training data for use in a deep learning convolutional network to detect multiple types of insects. In addition to labelling and training on images recorded under controlled conditions, also images were fed to the trained network that were recorded under uncontrolled conditions recorded with smartphone cameras.