Water & Wastewater Asia January/February 2022

SINGAPORE FOCUS
AUTOMATED MICRO-
INVERTEBRATE
DETECTOR
with image analytics
capabilities
By Lei Lei, Du Yu, Melissa Tay and
Nicholas Ho
Figure 1: Schematic diagram of the micro-invertebrate
detector that consists of multifilter-based concentrator
and optical detector
BACKGROUND
Regular operational monitoring of chironomid,
commonly known as midges, larvae, at
the water treatment plants is necessary to
detect for early signs of chironomid larvae
infestation. The current monitoring method
for chironomid larvae is very laborious as it
relies on plant operators to manually collect
samples, and a trained analyst to manually
identify and count live and dead chironomid
larvae under a microscope.
To reduce the time and manpower needed
for sample collection and analysis, PUB
and the NM3 Tech team jointly proposed
a new system to automate the detection
of chironomid larvae in water. This system
can detect chironomid larvae in 1,000 litres
of water sample within an hour with quasicontinuous
detection mode and minimal
maintenance requirements.
This solution will be trialled at selected PUB
installation, starting this year.
METHODOLOGY
Figure 1 shows the system structure of the
micro-invertebrate detector. In the multifilter
concentrator subsystem, a stainless-steel
filter of pore size 40μm is fixed on the filter
holder to concentrate the larvae in 1,000
litres of water. Thereafter, the trapped larvae
will be flushed down into a small volume
of concentrated sample by spray nozzles
positioned around the filter. The sample is
then transferred into the optical detector for
image capturing using a scanning imaging
system. The morphology of the trapped
particles and their movement information will
be recorded and analysed with a computing
processor comparing with the derived image
database.
In the optical detector, an initial coarse
scanning of the whole filter area using
0.5X magnification will be done to identify
the regions of interest. Subsequently, a
camera with higher magnification of 2X will
be positioned at the individual particle of
interest to capture detailed images for the
final artificial intelligence (AI) identification.
Furthermore, the viability is analysed based
on their mobility using image similarity
comparison method over the observation
period. The whole working process is
programmable and controlled by the system
software.
CURRENT ACHIEVEMENTS
The working prototype was installed at a
PUB installation for a three-month site trial.
The system was operating continuously
under 24/7 operation mode, and the system’s
No. of
Samples
Total
Spiked
Larvae
Results from blind spiked test samples
Manual Counts
(from captured
images)
Figure 2: The working procedures of the images
capturing, identification and viability analysis
stability and reliability were validated with
1,744 samples analysed on-site. The system’s
detection accuracy was validated with
151 blind test samples spiked with varying
numbers of chironomid larvae. An accuracy
of more than 80% was achieved for the
detection of chironomid larvae.
FUTURE PLANS
Phase 2 study is planned to complete a lab
version detector with multiple water sample
loader, while further improving detection
accuracy. In addition, the mechanical design
of the system will be improved for both indoor
and outdoor applications.
ACKNOWLEDGEMENT
This project was funded by PUB, Singapore’s
national water agency under R&D grant
(RND-Q4-43). We thank PUB for the financial
and technical support provided.
Lei Lei is CEO and Du Yu is scientific consultant at
NM3 Tech. Melissa Tay is senior biologist and Nicholas
Ho is biologist at PUB Water Quality Department.
Detector
Counts
% AI Accuracy
(Manual vs.
Detector
Counts)
% Accuracy
(Spiked vs.
Detector
Counts)
Blind test 151 292 265 235 88.7% 80.5%
18 WATER & WASTEWATER ASIA | JANUARY/FEBRUARY 2022