Building an artificial vision system of an agricultural robot based on the DarkNet system

Abstract

The article proposes a solution for the implementation of the artificial vision of an agricultural robot (farmbot). The Yolo3 system in the Darknet topology was chosen as the software platform. Recognition of objects (strawberries) and their discrimination with other objects in the farmbot workspace is provided by a convolutional network of deep learning with the primary use of the neuron activation function ReLU. The architecture used made it possible to ensure a percentage of correct recognition of objects up to 92 - 93% during dynamic processing of the video stream, including when processing objects that have a significant external similarity with the target object - strawberries. Similar results suggest that the farmbot computer vision system built on the Yolo3 platform in the Darknet topology will be fully operational.