Real-Time Instance Segmentation Techniques using Neural Networks for the Assessment of Green-Lipped Mussels

Mussel farming is a major industry in New Zealand that is currently growing fast. With the expansion of the industry, it has become important to allow the required processes to scale efficiently, which can be aided through the automation of tasks. One of these processes is harvest assessments, which...

Full description

Saved in:

Bibliographic Details
Published in:	2023 38th International Conference on Image and Vision Computing New Zealand (IVCNZ) pp. 1 - 6
Main Authors:	O'Keeffe, Hamish, Xue, Bing, Zhang, Mengjie, Hawes, Nicola, Lovell-Smith, Cris
Format:	Conference Proceeding
Language:	English
Published:	IEEE 29-11-2023
Subjects:	BlendMask CenterMask Instance segmentation Mask R-CNN Neural networks Performance evaluation Quantization (signal) Real-time systems shellfish Streaming media Task analysis YOLACT
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Mussel farming is a major industry in New Zealand that is currently growing fast. With the expansion of the industry, it has become important to allow the required processes to scale efficiently, which can be aided through the automation of tasks. One of these processes is harvest assessments, which are currently done manually by trained individuals who generally rely on their domain knowledge to perform the assessments rather than following a fixed standard. In this paper we propose the usage of real-time instance segmentation techniques to provide the basis for a tool that mussel farmers can use to automatically perform harvest assessments. Three neural network based real-time techniques were tested, YOLACT, CenterMask, and BlendMask, as well as Mask R-CNN. Mask R-CNN had the highest Mean Average Precision (AP) of 68.2 with a frame rate of 0.109 fps. CenterMask had the second highest AP at 63.7 with a frame rate of 0.278 fps. YOLACT had a final AP of 62.4 with a frame rate of 0.394. BlendMask had the lowest final AP at 60.9, with a frame rate of 0.354 fps. Though none of the techniques ran at real-time without an accelerator, the inference time for all three real-time methods was below 4 seconds. With the ability to run inferences on multiple mussels at a time, a non-real-time application using these techniques would still decrease the time taken to perform harvest assessments.
ISSN:	2151-2205
DOI:	10.1109/IVCNZ61134.2023.10343726