Work

A Dual-Stream Deep Learning Architecture for Action Recognition in Salmon from Underwater Video.

Year: 2017
Type: dissertation
Abstract: Over half of the costs from breeding salmon in the Norwegian salmon farming\nindustry comes from feed usage. Today the feeding process is largely a manual\nlabor, requiring an operator to monitor the amount of feed sinking to the bottom\nof a breeding cage. When the amount of feed exceeds a certain threshold, the\nfeeding process is terminated. Automation of this process and using salmon motion\nbehavior instead of sinking feed to determine when to terminate feeding, could\ngreatly reduce costs both in through the labor needed and amount of feed wasted.\nResent developments in Human Action Recognition have shown that Deep\nLearning approaches are well suited to perform Action Recognition. We\ntherefore examine the feasibility of using Deep Learning approaches to automate the\nfeeding process. We use 76 videos of salmon collected from within a breeding cage\nduring the month of November 2016. Using these videos for training, validation\nand testing, we propose three approaches to automatically classify Feeding and\nNonFeeding behavior in salmon. The three approaches are a Spatial Architecture, a\nSpatial Recurrent Architecture and a Dual-Stream Architecture.\nOur results show that all our proposed architectures are able to separate videos\nof Feeding and NonFeeding salmon with high accuracy. We also find that the Dual-\nStream Architecture is the best performing architecture. It combines spatial and\ntemporal information through the use of a Spatial Stream, a novel Temporal Stream\nand a Recurrent Neural Network (RNN). Our Dual-Stream Architecture is able\nto accurately classify 80.0% all of our testing videos, presenting state-of-the-art\nperformance.\nTo the best of our knowledge, both our Temporal Stream and our Dual-Stream\nArchitecture are original and novel architectures, as is the application of Deep Learning inference models for the Salmon Activity Recognition domain in optimization\nof feeding operation in Norwegian Aquaculture. We hope the results presented in\nthis thesis will contribute to achieve a higher sustainability in Norwegian salmon\naquaculture, optimize feeding operations, and consequently reduce potential waste.\nFuture work beyond the results presented in this thesis concerns research on understanding of what our the Deep Learning architecture have learned and visualizing\nthis learning process.
Language: English
Open Access status: green
OpenAlex ID: W2809657688
Cites:
Cited by:
Related to: 10