Home // ICAS 2020, The Sixteenth International Conference on Autonomic and Autonomous Systems // View article

Semantic Segmentation for the Estimation of Plant and Soil Parameters on Agricultural Machines

Authors:
Peter Riegler-Nurscher
Johann Prankl
Markus Vincze

Keywords: Semantic Segmentation, Agriculture

Abstract:
Many machine vision problems in agriculture, like plant classification, soil cover estimation or agronomic process evaluation in general, can be solved with semantic segmentation approaches. Naturally growing non-rigid organic and inorganic materials and plants are often characterized by blurred class transitions and high intra-class variance. Especially outdoor uncontrolled plant growth and plant decomposition lead to strong occlusions, cluttered scenes and strong illumination variances in images. An agricultural vision system has to cope with these challenges. This work presents four different applications for semantic segmentation in agriculture: (1) soil cover estimation, (2) estimation of grass-legumes ratio, (3) grassland swath detection and (4) grassland cut segmentation. For training, TensorFlow and a convolutional neural network are used. We investigate the influence of different pre-training methods to improve the overall classification performance with a limited number of training samples. The best test accuracy was achieved by initializing the weights from a model based on a semi-artificial clover and grass data set. The use cases with images from closer perspectives, (1) and (2), resulted in less accuracy compared to use cases (3) and (4). In general, all use cases can be solved with sufficient accuracy.

Pages: 87 to 90

Copyright: Copyright (c) IARIA, 2020

Publication date: September 27, 2020

Published in: conference

ISSN: 2308-3913

ISBN: 978-1-61208-787-0

Location: Lisbon, Portugal

Dates: from September 27, 2020 to October 1, 2020