Creating Fisheries of the Future

Our team is motivated to solve problems at scale and with global impact through computer vision approaches.  

End-to-end services

We build cloud-based systems to handle large-scale image and video data from sensitive marine conservation sources. We are passionate about user-centered design and intuitive web experiences catering to a wide range of users with built-in accessibility. Our hardware team love to push the boundaries of what is possible in Edge computing. Our AI experience straddles the gap between R&D and commercialization. With all these intersecting skillsets, we have an end-to-end service offering that can help your team realize innovative solutions for a sustainable world.

Our Process

We use an agile mindset and iterative delivery to realize your vision. Here’s an outline of what to expect working with us for bespoke projects.

Data & Discovery

In this activity we will examine sample data you may have available to inform our project design. We will learn about the problem you are aiming to solve.

Dataset & AI Development

In this activity we will collect and annotate imagery to realize our dataset design; and develop and train algorithm concepts.

User Experience &
Technical Design

In this activity we will create the look and feel of your application and design the infrastructure and/or hardware required to support it.

Solution Development

In this activity we build, integrate, and implement the solution.

Solution Pilot & Evaluation

In this activity we launch the solution for preliminary use and evaluate the impact.

In-House Annotation

The foundation of high-performing artificial intelligence algorithms is data. The foundation of good data is high-quality annotation. Ai.Fish provides annotation services that can support AI development or fine-tuning using video or still image data.
‍
Annotation involves two activities:

Localization of an Object

Depending on the AI approach that will be undertaken and the nature of the objects of interest, localization may be done a few ways including drawing a bounding box, setting a point, or segmenting the object.

Labeling of the Object

Labeling of an Object may include several different labels. For example, a fish might be labeled as a fish, but also as the species group it belongs to (e.g. tuna), and also as a detailed species (e.g. yellowfin tuna). This helps a dataset be useful for different types of AI training.

Custom AI Solutions

We specialize in applying computer vision to real world problems the ocean is facing. We are always looking for new partnerships and challenges that require specialized expertise in object detection, classification, or tracking, EdgeAI or custom AI augmented software applications.

Fisheries and Oceans Canada Proof of Concept: Automated detection of ghost gear (lobster traps and ropes) in side scan sonar imagery. Recovery of lost fishing gear (ghost gear) is a global problem impacting the safety of whales and other marine life in all ocean environments.

Got Data and wondering what you can do with it?

Send us a sample of your video or image data and a description of the problem you want to solve and we’ll let you know what we think is possible.

Not sure how to get started with AI but think it can help solve your problem?

Book a Discovery call with us and we’ll discuss your problem and potential solutions. During this call we can map out data needs to support AI strategies to inform your next steps.

Want to incorporate AI into your workflow but need a customized software support?

We can set up a consultation to understand your workflow and challenges and design software and AI supports that will integrate seamlessly.

AI Strategy

Introducing AI into your fishery can be a complex process. The success of the implementation often rests on the many choices that are made prior to the development of the AI. Here are some considerations you may want to examine in your fishery context.

Cameras

Many people ask us if there are minimum specifications or requirements for cameras? The truth is, there are a wide variety of cameras that can support AI. Important considerations include some of the settings you select on your camera, and how and where you mount it.

AI will generally work best with the following:
  • Minimum frame rate of 15 frames per second.
  • Minimum camera resolution of 1080p.
  • Overhead angles will work best for classification tasks (identifying species).
  • Cameras placed in locations that are optimized to minimize obstructed views when crew movement patterns are considered.

Fishery Features

The design and implementation of AI for fisheries is generally dependent on the gear used and the target and bycatch species expected. The technology is not at a stage yet where generic algorithms can apply to every context.

When considering the implementation of AI in your fishery we will benefit from answers to the following questions:
  • How consistent are the boat sizes and layouts?
  • How consistent are fishing practices from boat to boat?
  • How long is a typical fishing trip?
  • How much fishing effort is typical for each trip?
  • Is there seasonality to the target catch species or gear used?

Budget

AI implementations can vary widely with respect to cost. Custom solutions backed by novel algorithms will typically start at a proof of concept for a few hundred thousand dollars. Full implementations can require a 2x or 3x investment.

Key costs to consider include:
  • How long you will store your data?
  • How much data you need to store?
  • How accurate your AI results must be?
  • How quickly your AI results are needed?

Web-based experiences

In a world where there’s always “an app for that”;
we are bringing modern software experiences to commercial fishing.

User Experience
User Interface
Software Development

User experience, often referred to as UX, considers all the elements that shape a user’s ability to accomplish tasks in a software product or service. A user experience designer will consider the software elements, how these make the user feel, and the ability of the user to know what to do next during any given interaction. Our in-house UX team will consider your fishery context, current and ideal user workflows, desired time spent on interactions, and the easiest path to achieve your outcome during experience design. Often the output of this first design stage will include a written use case document that describes the ideal user workflow and context for each step. Following user experience design we proceed to User Interfaces.

A user interface design presents the visual design of all the software elements required to deliver the designed user experience. In web-based software this includes the screen display on the target device including elements like color and typography. It includes interactive elements such as buttons, menus, form fields, navigation from one workflow step to the next, and the way data are presented to a user if any user feedback is required during the experience. Finally, a good user interface design will also consider visual cues that help a new user easily execute the desired workflow and remind experienced users to stay consistently within a workflow. These interface design elements may include animations, hover states, and scrolling or pagination. User interface designs may go through several iterations starting from basic designs showing only the elements of the software - called wireframes, and then progressing into higher fidelity designs that feature full-color screens and simulated interactions also known as “clickable prototypes”. After we’ve aligned on user interface designs we develop the software.

At Ai.Fish we develop responsive web-based applications. This means that our applications can be used and viewed on different screen sizes served only from one codebase. Customers can choose to host in their preferred cloud environment or within the Ai.Fish cloud. We also design systems that serve the web application from a locally hosted server. Bespoke software development can incorporate your brand elements and guideline. Interfaces can be multi-lingual and can be designed to suit your desired accessibility criteria.

Cloud-based systems

Our full-stack development team builds cloud-based systems that process large-scale image and video data while maintaining strict confidentiality and compliance with applicable privacy regulations. This team includes dedicated machine learning operations (MLOps) and expertise in the latest cloud and web-based tooling including Django, Flask, React, Vue, Docker, Kubernetes, Node, Terraform, and TypeScript. Our team is comfortable working in any of the major cloud environments. We can design implementations that are hosted in your cloud of choice or hosted for you in our Microsoft Azure environment. Developing software for the cloud is a distinct skillset compared with traditional enterprise software implementation. Here are some things to expect in a cloud-based system:
Data Storage & Security

AI applications rely heavily on the image or video data they will act upon. This means that data storage must be accessible to AI processing. Encryption and decryption of stored data must happen rapidly. The private nature of many image and video sources in sensitive marine contexts necessitates encrypting data not only while at rest but also during transit. The data architecture must ensure data are segregated between system tenants if the system will be multi-tenant. We designed redundant cloud-based systems, meaning data are automatically backed up to a location outside the primary storage to ensure continuity in the event of a system failure. Data access is typically controlled by Application Programming Interfaces (APIs) between end-user applications and data as well as between service applications and data. This ensures data security through the passing of required secure credentials to facilitate API access.

AI Processing

AI applications are compute intensive operations, sometimes requiring costly server infrastructure in the cloud. We design applications that are scalable to ensure costs are optimized. This means that the system will automatically scale-up when data is arriving, and scale back down when processing is complete.

Microservice Design

We favor a microservice design approach to cloud-based systems. This means we design small components that perform a specific task and are loosely coupled to each other or functional within an expected system flow. Microservices design maintains flexibility and agility within large systems and can help to reduce ongoing development costs and maintenance cycles.

Continuous Integration/Continuous Deployment

We typically deploy cloud-based systems that are supported by code repositories configured for continuous integration and continuous deployment. This means interface updates, API updates, and service configuration management, can be released seamlessly without disruption to end users.

Edge Based Systems

Edge-based systems bring the opportunity for real-time AI-assisted data analysis at sea, eliminating the delays and connectivity requirements of cloud-based processing.  Edge computing benefits include:

  • Real-time results available onboard to support fishing decisions that maintain sustainability and compliance
  • Real-time results that provide easier and more accurate logging for fishermen
  • Real-time results that ensure shore-based compliance teams can respond to incidents rapidly
  • Reduced data storage and transfer burdens if desired

Our Edge hardware team stays current on the latest and greatest technologies available from various suppliers of Edge system components such as compute devices, connectivity, and camera options. We develop our own Edge operating systems to optimize compute capacity for AI processing.