Article
Weed Classification Using Explainable Multi-Resolution
Slot Attention
Sadaf Farkhani * , Søren Kelstrup Skovsen , Mads Dyrmann , Rasmus Nyholm Jørgensen
and Henrik Karstoft
Citation: Farkhani, S.; Skovsen, S.K.;
Dyrmann, M.; Jørgensen, R.N.;
Karstoft, H. Weed Classification
Using Explainable Multi-Resolution
Slot Attention. Sensors 2021, 21, 6705.
https://doi.org/10.3390/s21206705
Academic Editors: Asim Biswas,
Dionysis Bochtis and Aristotelis C.
Tagarakis
Received: 30 June 2021
Accepted: 1 October 2021
Published: 9 October 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Department of Electrical and Computer Engineering, Aarhus University, 8200 Aarhus, Denmark;
ssk@ece.au.dk (S.K.S.); madsdyrmann@ece.au.dk (M.D.); rnj@agrointelli.com (R.N.J.); hka@ece.au.dk (H.K.)
* Correspondence: farkhanis@ece.au.dk
Abstract:
In agriculture, explainable deep neural networks (DNNs) can be used to pinpoint the
discriminative part of weeds for an imagery classification task, albeit at a low resolution, to control
the weed population. This paper proposes the use of a multi-layer attention procedure based on a
transformer combined with a fusion rule to present an interpretation of the DNN decision through a
high-resolution attention map. The fusion rule is a weighted average method that is used to combine
attention maps from different layers based on saliency. Attention maps with an explanation for
why a weed is or is not classified as a certain class help agronomists to shape the high-resolution
weed identification keys (WIK) that the model perceives. The model is trained and evaluated on two
agricultural datasets that contain plants grown under different conditions: the Plant Seedlings Dataset
(PSD) and the Open Plant Phenotyping Dataset (OPPD). The model represents attention maps with
highlighted requirements and information about misclassification to enable cross-dataset evaluations.
State-of-the-art comparisons represent classification developments after applying attention maps.
Average accuracies of 95.42% and 96% are gained for the negative and positive explanations of the
PSD test sets, respectively. In OPPD evaluations, accuracies of 97.78% and 97.83% are obtained for
negative and positive explanations, respectively. The visual comparison between attention maps also
shows high-resolution information.
Keywords:
transformer; slot attention; explainable neural network; fusion rule; weed classification;
weed identification key; precision agriculture
1. Introduction
Weeds compete with crops to capture sunlight and take up nutrients and water; this
competition leads to significant yield losses around the world every year [
1
]. Furthermore,
there are considerable indirect negative externalities that should be taken into consideration
when combating weeds [
2
]. Currently, the use of conventional weed control methods
usually results in soil erosion, global warming, and human health problems [
3
–
6
]. Weeds
are usually not distributed evenly across farmlands. Therefore, weed management could
be greatly improved by collecting information about the location, type, and amount of
weeds in an area [7].
In general, there are three primary weed management strategies: biological, chemical,
and physical [
8
]. Biological weed management refers to weed control through the use
of other organisms, such as insects or bacteria, to maintain weed populations at a lower
level [
9
]. Biological weed control is, however, a prolonged procedure that reduces the
growth of a specific species. Selective chemical weed management using an autonomous
and unmanned vehicle is one solution for controlling the weed population and requires the
use of considerably lower contamination doses [
10
]. In the physical approach, weeds are
controlled without herbicide; this is typically accomplished through the use of mechanical
tools. Physical weed control requires extra precision in the detection of weeds, as non-
selective and incorrect weed detection can harm the crop.
Sensors 2021, 21, 6705. https://doi.org/10.3390/s21206705 https://www.mdpi.com/journal/sensors
