Meistaravörn við tölvunarfræðideild: Elías Ingi Elíasson

A Convolutional Neural Network Architecture Benchmarking for Semantic Segmentation of Fish Fillet Images

  • 3.6.2019, 10:00 - 11:00

Student: Elías Ingi Elíasson
Title: : A Convolutional Neural Network Architecture Benchmarking for Semantic Segmentation of Fish Fillet Images
Date and Location: June 3rd 2019 at 10:00 in room V102
Supervisor: Stephan Schiffel, Assistant Professor, Department of Computer Science

Abstract: There are many factors that have led to increased demand for automation in the food processing industry, such as labor cost, work safety and improved yield of the product. Many of the current algorithms for cut patterns and pin bone removal in fish require multiple engineering hours as they have to be custom fitted to each different species.

Utilizing machine learning methods for the task has been a recent presentiment, as it offers an agile solution to adapting to different fish species, and is already showing promising results. However many such models, namely convolutional neural networks, are often focused on achieving new heights in accuracy and may be too intensive on computational resources to be able to operate in systems within the industry. We research recent state-of-art models for semantic segmentation, and benchmark three candidate models for the task on detection performance and several metrics regarding computation. We use 396 RGB cod fillet images with corresponding target masks representing six different classes, for evaluating pixel-wise model predictions and computational performance. U-net and FC-DenseNet56 performed best on detection performance scoring a mIoU of ∼ 59%, while with E-net we report a fourfold reduction on inference, compared to the other two, reduced training time and the capabilities to run on a much smaller GPU, at the cost of ∼ 4% mIoU.



Vinsamlegast athugið að á viðburðum Háskólans í Reykjavík (HR) eru teknar ljósmyndir og myndbönd sem notuð eru í markaðsstarfi HR. Hægt er að nálgast frekari upplýsingar á ru.is eða með því að senda tölvupóst á netfangið: personuvernd@ru.is
//
Please note that at events hosted at Reykjavik University (RU), photographs and videos are taken which might be used for RU marketing purposes. Read more about this on our ru.is or send an e-mail: personuvernd@ru.is