A proof of concept project collaboration with A*Star (IHPC) and (NMC) to develop a feasibility study to assess potential of acoustic-based sonar technology for RAS biomass estimation in the Mixotrophic System.

Problem Statement

• Need for solution for biomass estimation
• Manual, labour-intensive task
• Imposes stress on fish
• Mixotrophic system increasing water turbidity
• Intensive and super-intensive RAS:
  • Occlusions
  • Aeration

 

Objectives

Develop a feasibility study to assess potential of acoustic-based sonar technology for RAS biomass estimation.

Project Components

A)Experimental Design for Sonar Imaging of Fish in Aquatic Tank with Murky Water Conditions
B)Functional Test Sonar Images for Aquatic Tank with Murky Water Conditions
C)Aquaculture Biomass Estimation Software based on Sonar Images for Murky Water Conditions

Experimental Design for Sonar Imaging of Fish in Aquatic Tank with Murky Water Conditions

NMC designed the experimental setup of the underwater POC for sonar imaging of fish including device selection, mounting design and device placement evaluation.

Further refinements to the experimental setup were conducted throughout the project duration from the lessons learned in performing the experimental data collection and the data analysis and biomass estimation.

NMC and IHPC has also jointly defined the set of individual experimental conditions for data collection, taking into consideration performance specs of the sonar device including frequency range and field-of-view as well as the experimental setup, and the data needed to perform the biomass estimation.

Functional Test Sonar Images for Aquatic Tank with Murky Water Conditions

NMC evaluated three different models of sonar devices at 4 different periods including Tritech Gemini 1200ik (Jul), Blueprint Oculus M1200d (Sep) and Sound Metrics ARIS Explorer 3000 Sonar (Aug and Oct).

The experimental requirements/conditions were changed as the optimization of experimental model design in stages;

Tritech Gemini 1200ik (Jul 2020)

184 experimental cases with different configurations and settings. Experimental configuration variations: Optimizing sonar position and its depth to different sonar gain and frequencies.

(a) 6m tank with 4 aeration, 100% gain and 100 small fishes in clear water; (b) 3m tank with no aeration, 100% gain and 100 big fishes in murky water; (c) 3m tank with 1 aeration, 100% gain and 50 small fishes in clear water; (d) 6m tank with no aeration, 100% gain and 100 big fishes in murky water.

Sound Metrics ARIS Explorer 3000 (Aug 2020):

Experiments conducted with a higher frequency range up to 3 MHz.

The Sound Metrics was evaluated by optimising the experimental configurations.

a) 6m tank with 4 aeration, 1.8 MHz with max gain and 100 kg fishes in clear water; (b) 6m tank with 4 aeration, 1.8 MHz with max gain and 32 kg fishes in murky water; (c) 6m tank with 4 aeration, 1.8 MHz with max gain and 100 kg fishes in murky water; (d) 6m tank with 4 aeration, 3 MHz with max gain and 100 kg fishes in murky water.

Blueprint Oculus M1200d (Sep 2020):

A fishing net at the centre to create a more controlled volume so that fishes in a more restricted vicinity. Blueprint could not meet our requirements. There were no visibilities of fishes and too much background noise despite clear water conditions and blank images in higher frequency.
With almost the same experimental condition but in a very murky water, Tritech was able to measure and identify fishes, producing good quality fish images.

Comparison of images by (a) Blueprint in clear water and (b) Tritech in murky water, both in 6m tank with 10 fishes

Sound Metrics ARIS Explorer 3000 (Aug 2020):

Controlled volume with the aid of a fishing net so that fishes are within the field-of-view (FOV) of the device. To further assist the labelling and model training process, a GoPro Camera was integrated with the sonar device together with a torch light. The total number of fishes measured by Sonar device and Camera show good agreement in clear water conditions.
It can also be deduced that this Sonar is able to detect and capture fishes in murky water whereas cameras are not suitable to measure fishes in murky water. Sound Metrics Sonar device was able to produce good quality sonar images compared to other models

Comparison of images by Sound Metrics in (a) clear water and (b) murky water.

Image Processing and Biomass Estimation

Dataset Summary:

• Experiment 1: Gemini 1200ik, Tritech (720 kHz, 1200kHz)
• Experiment 2: (ARIS 1) Sound Metrics ARIS Explorer 3000 (3 MHz)
  • 50 cases, 0 to 102.9 Kg
  • 25 clear water, 56 murky water
  • The fish are free in the whole tank
• Experiment 3: (ARIS 2) Sound Metrics ARIS Explorer 3000 (3 MHz)
  • 50 cases, 0 to 50.79 kg,
  • 24 clear water, 26 murky water
  • Fish restrained to be in sonar field of view as much as possible

Machine Vision Algorithms and Biomass Estimation:

Conclusions (Tritech):

• Biomass estimation not consistent due to noise
• However, established feasibility of sonar for imaging in RAS tank

 

Conclusions for Sound Metrics (Round 1):

• Image quality generally better (less noise) than Tritech
• Biomass estimation consistent, but error is high
• Issue with field of view/tank coverage

 

Conclusions for Sound Metrics (Round 2):

• Maximum achieved accuracy is estimated to be 90.8%
• RMSE In biomass estimation: 4.69 kg

 

Key Conclusions

Technical feasibility of sonar for biomass estimation established. Our developed AI machine vision algorithms have good performance. There is a lack of complete field of view, tank coverage are limiting factors which needs to be addressed for future implementations.

SHARE THIS ARTICLE

---