Measurement of growing situation of agricultural crops on FPGA-mounted drone using Circle SSD

Abstract

In this paper, we report the optimization method for the case of introducing SSD which is one of the object detection models to the drones for the purpose of measuring the growth situation of agricultural crops as viewed from a bird's eye viewpoint. We have adopted a “Separable Convolution” and introduced “Circle SSD” which uses circular shape as a detection frame as an optimization method. As a result of the optimization, the F-measure 0.67 was realized with 14770 convolution params, and the object detection could be realized with limited calculation resources on the drone.