Pest detection systems and methods

Abstract

A pest 6 detection system (1, Fig.1), comprising: a processing unit configured to receive image data and use a machine learning model to determine one or more pest parameters. The parameters are recorded on a computer readable medium (12, Fig.1). The pest detection system may include a soil tiller (50, 51, Fig.5), the camera (111, Fig.5) capturing images of the agitated soil. The pest parameters may include presence or absence of a pest, a number of pests and / or type of pest. The machine learning model may be trained using training images of a proxy for a pest; such as a mealworm substituting for a wireworm. The training images using may be divided into tiles (a-d, Fig.19) and the tiles (c, Fig.20) used for training. A captured image may be subject to a sliding window that divides the image into overlapping tiles. The pixels in each tile may be assigned confidence value that the pixel shows a pest. The confidence values for each pixel may be averaged. The number of pests may be determined by binarizing the confidence values for each pixel with a threshold value, determining the coordinates of localised confidence peaks and counting the number of localised peaks.

Images