OBJECTIVE OF THE PROJECT

To create a SUMMARY that provides a complete overview of the project.

The Integrated Natural Environment Management System (SigMedNat) maintains a database with relevant information from the General Directorate of Natural Environment of the Government of Cantabria. This system allows for the visualization and categorization of contacts with the natural environment, consisting of photo captures (images and videos) obtained through a network of automatic shooting cameras distributed in grids across the southwestern area of Cantabria.

PROJECT CO-FUNDING

Project co-financed up to 50% of the operation’s value by the General Directorate of Innovation, Technological Development, and Industrial Entrepreneurship of the Ministry of Innovation, Industry, Tourism, and Commerce of the Government of Cantabria, and by the European Union with resources from the European Regional Development Fund (ERDF) through the ERDF Operational Program 2014-2020 of Cantabria.
Both the identification of species and the categorization, along with other relevant aspects such as health status, are determined by one or more experts who review the photo captures. The database of this system has the following characteristics:

1. The system covers 12 regions: Liébana, Nansa, Saja, Campoo, Valderredible, Besaya, Pas, Miera, Asón, Eastern Coast, Central Coast, and Western Coast.
2. A total of 84 grids are distributed across the regions covering the area.
3. Among the grids (mainly in the southwestern area), a total of 93 cameras capture frames of the natural environment.
4. Captures are categorized based on the species of animals present, allowing for a total of 878 different species.
5. The species are organized into 6 distinct classes: Amphibians and reptiles, Birds, Mammals, Bats, Freshwater fish, and Others.
6. Captures taken by the cameras are also organized into various projects according to the monitored animals: Wolf, Brown Bear, Wild Boar, Fox, Roe Deer, Deer, Livestock, Rodents, Birds, and Domestic animals. Additionally, categories for Other species of interest, Mobile cameras, and Unidentified species are included.
7. Each contact recorded by a camera can contain multiple images and/or videos.
8. For each photo capture, efforts are made to identify the species present in the image, as well as the number of animals present by sex (male, female, or unknown) and age (juveniles, subadults, and adults).
9. Captures corresponding to humans are discarded from the system.
10. The photo captures present very diverse conditions of climate, lighting, environment, and image quality.

TYPE OF PROJECT

Industrial Research
The main objective of the project is to specify, design, and implement an experimental laboratory prototype that allows for the automatic identification and categorization of animal photo captures obtained through the previously mentioned network of shooting cameras, as well as facilitate species monitoring for expert biologists.
Specific objectives include:

Objective 1. Build an experimental prototype in an environment with simulated interfaces that allows for the identification and categorization (class, species, number, age, sex) of animals captured by the cameras using data mining and machine learning techniques, specifically deep learning, thus replacing human intervention in the information processing activity. This prototype must consider different conditions of lighting, climate, environment, and resolution.

Objective 2. To better manage wildlife in general and protected species, the system must assist in achieving the following sub-objectives:

• 1. Determine the relative proportion of species.
• 2. Infer the presence or absence of species.
• 3. Establish a pattern of activities for each species, if possible.
• 4. Determine distribution by grids and regions.
• 5. Emphasize information on bears, wolves, and wild boars.

Objective 3. The system will use a deep learning methodology, and the fundamental categorization algorithm must be automatically or semi-automatically reconfigurable as the database with labeled frames grows. This approach aligns with cutting-edge machine learning ideas, where the architecture of computational models ideally should be updated automatically over time, constituting one of the significant challenges in machine learning research.

Objective 4. Animal monitoring will be enabled through the use of probabilistic models considering species distribution and their movement between different photo capture areas. Observational learning techniques will be used to extract the behavioral model of the species.

The results of the project development will allow Axpe Consulting Cantabria to offer solutions to companies dedicated to the management of the Natural Environment of Cantabria, which require services for the identification and recognition of animal images taken through surveillance cameras.

Advancements in data processing techniques and machine learning, primarily based on deep learning, particularly convolutional neural networks, will allow Axpe Consulting Cantabria to position itself as a leading company in the field of Artificial Intelligence, which is currently strategic for the company. In the automatic identification and interpretation solutions of images and videos, which are currently in high demand due to the needs of tracking the COVID-19 disease.

PRODUCT, PROCESS, AND/OR SERVICE

Detailed explanation of the novel product, process, and/or service that could become a business reality.
To conduct this study, the construction of a prototype in a laboratory environment is proposed to experiment with species categorization. This prototype will consist of the following components that will operate in combination:

• The experimental prototype of an application that allows capturing and processing the movement data of animals stored in the database.
• Computational models to represent the movements.
• Software based on automatically learned convolutional neural networks that allows categorizing species and other required attributes.
• Generic web services to communicate the predictions of the convolutional neural network interoperably to systems managed by the biologists responsible for the program.
  The experimental study will evaluate the following aspects to determine the viability of the proposal for future evolution into a marketable product usable by the biologists in charge of the program:
• A study with frames that will determine the success rate for correctly categorizing species. This success rate will be quantified as the percentage of agreement between the results of the convolutional neural network and the categorizations made by experts.
• An evaluation according to the ISO/IEC 25000 standards family, known as SQuaRE (System and Software Quality Requirements and Evaluation), which will allow for standardized measurement of compliance with essential functional and non-functional requirements in the infrastructure of the experimental laboratory (security, performance, reliability, etc.).
• An evaluation by experts of the impact of the proposal on species categorization.

LOCATION OF THE PROJECT

Axpe Consulting Cantabria S.L.
Polígono Camargo 16C
39600 Camargo, Cantabria

Project Start Date
Always after or equal to the date of application.
December 1, 2020

Project End Date
At most August 31, 2022.
August 31, 2022.”