Species Detection With Machine Learning: Simplifying Efforts for Conservation Organizations

Machine learning has evolved its use cases in Species detection. This has resulted in making conservation researchers, and non-profit organizations self-sufficient to use data and technology to save species. Be it saving the declining population of penguins in Antarctica or Sumatran Rhinos that are on the verge of extinction, technology is serving as a helping hand.

Now, they can utilize data to its full extent and derive actionable insights by building machine learning models to detect any kind of species.

Excessive hunting, poaching, habitat destruction, pollution, and human-wildlife conflicts, apart from some natural causes, have endangered so many species across the globe.

Additionally, rampant deforestation in the name of land development and farming is leading species to extinction.

What is Species Detection?

So, what if a few species go extinct?

For starters, let’s take the example of the tiny bee.

Despite over 20,000 species of the bee, it has been officially included in the endangered list. What happens if the bees go extinct? Well, humans may go extinct soon after.

Southwest Wildlife Trusts’ 2019 report mentions an “unnoticed apocalypse” that has possibly killed 50% of the world’s insects.

The disappearance of insects, in turn, harms the flora and fauna and the entire ecosystem. The report points out some ways in which this decline can be tackled:

• Stop using pesticides routinely and unnecessarily
• Start building nature recovery networks — gardens, parks, green balconies & terraces, etc.
• Help in making farms more wildlife-friendly and sustainable
• Have more robust legal frameworks for insect and wildlife protection
• Give protected status to more extensive and more areas with semi-natural habitat

BBC gives suggestions to people in urban areas about how they can support insects and birds:

• Growing plants in whatever little space they have in balconies or elsewhere
• Helping bugs sip water — keep water in shallow trays with some pebbles
• Making homes for bees
• Eating organic fruits and vegetables
• Keeping outside lights off or minimal for as long as possible

The above suggestions can help in mitigating the decline or probably resurrecting specific endangered species populations too. However, it is also essential to discover what new species might have already become or are on the verge of becoming endangered or extinct.

The sheer number and diversity of species present a significant challenge to species detection.

As per a 2011 study published in PLOS Biology, there are an astonishing 8.7 million species on land and sea combined. Of these, only about 1.2 million have been identified and cataloged.

A staggering 7.5 million are still to be identified. The challenge is daunting. Let’s see why.

Challenges Researchers and Non-profits Face in Identifying Species in the Wild

To begin with, it may not be easy to identify rare species because they are rare! Some of them may additionally be elusive or nocturnal or be extremely difficult to spot.

Moreover, when researchers engage with those species in their research attempts, they may endanger the species through human contact or by inadvertently allowing accessibility to poachers.

Expensive Technology

Scientists and researchers are now acknowledging new-age technology as well.

For example, Stuart Pimm, a conservation ecologist from Duke University, places camera traps with digital cameras in remote places and tracks movements of elusive species like snow leopards and tigers.

He also talks about using high-resolution drone cameras. Though the technology used is expensive, combined with the right AI models, it can give results that would outweigh the costs.

Another challenge faced by scientists and researchers is studying animals that move in groups.

It is difficult to count and observe multiple animals at one go with human eyes. It is these limitations that scientists are now looking to overcome by using technological advancements.

They hope to enhance the scale of inquiry, the volume and quality of data, and data processing speed.

Intensive Time and Effort

Some scientists and researchers are looking at more innovative means of finding or researching different species.

For example, Emma Bennett, a wildlife ecologist from Melbourne, has found an innovative alternative to traps that can harm animals — she is using trained dogs to detect tiger quoll’s feces which can be analyzed to understand the sex, diet, and distribution of the species.

But we can easily see the time, effort, and safety concerns of this method.

Lack of Analytical Skills

Another problem scientists face is the analysis of raw data. Even though advanced technology such as high-resolution drone cameras may give copious data, scientists may find it challenging to put the data to use.

Deep-learning-based AI solutions to conserve biodiversity can come to the rescue here.

One of the prime examples is the Nisqually River Foundation working towards identifying salmon species in the Nisqually River. The Washington-based non-profit organization was counting fishes manually after capturing their images through underwater camera traps.

This led to weeks of manual efforts from scientists to build fish species classification repository.

Gramener helped them organize their image data and build a Salmon Detection Web App on top of it. The machine learning-driven fish species detection web app has automated the fish counting for them.

Now, biologists can dedicate their time to many important tasks rather than sitting and counting fish manually. The innovation resulted in 80% potential cost-saving reducing the manual efforts by 5 times.

Lack of Crowd-Sourced Data on Species

Paul Evangelista, a Colorado State University research ecologist, surveyed the indigenous citizens of Somaliland, a region of Somalia, in 2016–17 and enquired whether they had spotted any of the 25 species on his list.

Some extinct species were included for quality control in case of false reporting. This model helped Evangelista’s team in building animal distribution models.

While this attempt is an excellent example of crowd-sourced data, not everyone can use such methods due to social, ecological, and accessibility concerns.

Lack of Insights-Driven or Data-Driven Approach to Implement and Monitor Schemes

Scientists and researchers are experts in their field, but they may not be trained to handle large amounts of data that is now possible because of advanced technology.

Till now, researchers have primarily relied on observational studies instead of experimental evaluations.

However, with new technology — computer vision, machine learning, deep learning — it is possible to get the help of mathematical and computational models in understanding phenomena like coordinated motion and collective predator evasion.

Can We Automate Species Detection?

Species detection can be daunting, especially in the case of dense habitats and fossorial animals.

But different methods such as low-resolution airborne thermal imagery, autonomous underwater vehicles (AUV), and camera traps are being used to collect images and videos that share enormous data about the lives of hitherto elusive species.

Of these automated methods, species identification through camera traps is increasingly used by biologists, researchers, conservationists, and even hobbyists because the camera traps are now becoming affordable.

The camera trap is a device that contains a digital camera connected to an infrared sensor. When a warm object such as any animal moves close to the infrared sensor, it starts the camera and records information until it is out of range.

These camera traps can be left to record data for weeks, even months. This allows having long-term records of animal behaviors.

Camera traps are also “wildlife-friendly” as they cause minimal intrusion into wildlife. Moreover, the data collected is permanent and verifiable.

Data about species location, species population, and species interaction is collected. Networked camera traps have the capability of sending information over the phone or satellite networks in almost real-time.

This is extremely useful for curtailing poaching and hunting practices.

The copious data that is collected through these camera traps can be trained to identify images of species through artificial intelligence — machine learning to be specific.

In species detection, computer vision, machine learning, and deep learning software use algorithms and statistical models to train computer systems to classify, categorize, and identify images.

Once the system is trained, it can organize and identify thousands of pictures in seconds, a feat impossible for humans to achieve.

ML-Driven Species Detection API

Gramener has created a solution for species detection under Microsoft’s AI for Earth initiative. The API or application program interface used for species detection runs using AI for Earth API backend.

Try Demo

This AI-ML-based species detection tool can help in biodiversity conservation. It can help to detect, identify, and monitor all types of flora and fauna species. The solution can help tackle the challenge of bottlenecks created by delays in identifying images collected by scientists.

It uses Deep Learning Virtual Machine, in which deep neural networks are trained to identify flora and fauna species in images.

The model uses input images of plants and animals. These images are uploaded to the API, which supports more than 5000 species.

When run on the API, all candidate species resembling the input image are returned with a matching percentage called confidence value.

Both common and scientific names are returned. If needed, bounding boxes are also included for different species. It is a ready-to-deploy solution for organizations that aim to identify species with advanced analytics and Machine Learning models.

Try our subscription-based species detection solution that comes with:

• Zero investment
• Ready-to-use API plugin
• Ready-to-deploy solution
• Easy integration with all platforms
• Subscription-based pricing — pay as per usage
• Rapid customization and training of model to detect new species of interest

Why is Species Identification Important?

The Earth is teeming with life and knowing how many species inhabit the world is a fundamental but elusive science question.

Learning about geographic distribution and species evolution is vital for biodiversity conservation and humanity's sustainable development.

Here are some other essential uses of species identification:

Monitoring of Population Trends

It is crucial to monitor species and their populations over long periods. Any drastic change can adversely affect the region’s ecological balance, the planet, and human life.

Implementation and Evaluation of Population Management Programs

Programs meant for population management of specific species must be appropriately implemented and evaluated.

Properly implemented programs can help in identifying any significant drop or rise in the particular population.

This can help with timely intervention. For example, population management may be crucial in the case of invasive species. Such species are non-native and can harm the ecology, environment, or even human health.

Health Assessments of Ecosystems

Identification of species and studying their populations can also help assess the health of ecosystems.

A deteriorating ecosystem will harm biodiversity and thereby hurt human health. A loss of ecosystem vigor because of fragmentation of landscape can lead to a decline in ecosystem health.

A change happening even in coastal regions can have an impact on the inland cities and counties.

Extinction Analysis

Species identification can also help in extinction analysis. Knowing the rate of population decline in endangered species can help scientists and researchers determine the time of extinction of such species and its impact on the human population.

It also helps determine at what rate currently safe species populations can become threatened or even extinct.

The Negative Impact of Species Misidentification

Correct wild animal identification is essential in species detection. Otherwise, it can lead to the following negative impacts:

Accidental Culling of Endangered Species

Endangered species must be protected to prevent harmful effects on the ecosystem and environment.

Misidentification of the species can lead to the involuntary culling of endangered species, which can be dangerous for the ecosystem.

Incorrect Monitoring of Harmful Algal Blooms

If incorrectly identified or monitored, harmful algal blooms may cause mortality of fish masses, poisoning of humans, and economic loss.

It is essential to monitor such harmful algal blooms to understand their nature, biology, toxin chemistry, and ecophysiology.

The Unobserved Decline in Significant Fish Stocks

The genetic and behavioral diversity of fish stocks is often overlooked. This happens because it is common to study models of several distinct stocks as one large stock.

This creates a false growth and harvest potential leading to sudden collapses even in the so-called regulated fish stocks like cod.

The generalization of different stores as a single large stock happens because of a lack of appropriate tools.

Drafting Inappropriate Management Plans From False Species Sightings

Human error in identifying the correct species is documented. It compares for both experts and non-experts, with both groups scoring a 60% accuracy.

It can be more pronounced for concurrent members of the same or different species.

While it is common practice to take the help of native citizens in scientific research, such assistance can be wrought with inaccuracy. Moreover, it can lead to wasted effort if management plans are created for the wrong species.

Conclusion

Conservation of biodiversity is essential and is heavily dependent on accurate species identification.

It is difficult to rely on human effort to have accurate ecological data, especially when the number of species runs in millions.

Advanced technology such as artificial intelligence, machine learning, deep learning, and computer vision can help identify species quickly and accurately.

Gramener’s AI-ML-based species detection solution can help scientists and researchers greatly in this effort.