Impact Spotlights: AI in the Wild Highlights Innovations for Environments and their Inhabitants

Hackster's Impact Spotlights: AI in the Wild livestream featured a lineup of talented engineers and developers who utilized edge AI, machine learning, and low-power IoT hardware to tackle multiple environmental challenges. From early detection of algae blooms to safeguarding endangered elephants, the projects highlighted how AI is being used to monitor, classify, and respond to real-world issues in real-time.

Smart Lake

The first guest speaker, Sashrika Das, presented her innovative Smart Lake device that employs sensors and machine learning at the edge to detect harmful algal blooms (HABs) in freshwater environments. Once detected, the device will notify residents or authorities in real time to prevent accidents.

Das designed the Smart Lake device using a Wio Terminal development board that's driven by an ATSAMD51 and comes equipped with LoRaWAN to transmit data via a Wio LoRa Chassis to Helium Network hotspots. The device also packs a host of sensors, including a pH sensor kit, turbidity sensor, temperature and sunlight sensors, all of which are utilized to monitor fresh water.

Das also developed an "eventer" API gateway that allows external systems (e.g. public health, town halls) to subscribe via callback URLs and receive push notifications if algae blooms are detected.

Remote Birding with Edge AI and Blues

The second guest speaker, Rob Lauer, was on hand to present his Remote Birding platform, which takes advantage of TensorFlow and Blues Notehub to identify different birds. Lauer designed the platform around the Raspberry Pi 4 Model B along with a Pi Camera, PIR motion sensor, Blues Notecard and a Notecarrier-Pi, which provides cellular and GPS connectivity.

Taking into consideration how much power the platform would consume in the wild, Lauer outfitted the device with a BigBlue portable solar charger (42W) and a ROMOSS 30000mAh power bank. On the software side, the Remote Birding system runs a pre-trained TensorFlow Lite bird classification model, with captured images classified locally and transmitted over a cellular network via the Notecard.

AI-Powered Radio Collar

The third guest, Rucksikaa Raajkumar, detailed her innovative AI-powered radio collar designed to prevent human-elephant conflicts and poaching, which are major threats to the endangered elephant population in regions like Sri Lanka. Raajkumar's novel approach takes advantage of RFID, LoRaWAN, GPS tracking, and the Avnet IoTConnect platform, which enables real-time tracking, alerting, and human identification to keep elephants and communities safe.

Raajkumar designed the collar around a SparkFun Simultaneous RFID Reader (M6E Nano) equipped with a UHF RFID antenna and a Semtech LR1110 LoRa transceiver. Each Elephant has a unique RFID ID that allows for individual tracking. Simultaneous RFID readers and ultra-high frequency antennas detect RFID tags, while RFID readers estimate the distance of an elephant from communal or high-risk areas.

Passive RFID tags are also issued to all authorized personnel (rangers, safari-goers, law enforcement, etc.), allowing RFID systems to identify authorized personnel or unwanted poachers. If unauthorized individuals are detected near the elephants, the system sends out an alert.

Raajkumar is also utilizing microphones embedded in the collars to identify if predators or poachers are near based on their unique calls. To train her machine learning model, Raajkumar tapped data from the Elephant Voices database, which acts as a repository of elephant sounds that were garnered over decades.

IoT AI-Driven Tree Disease Identifier

The session's fourth guest speaker, Kutluhan Aktar, showcased his IoT AI-driven Tree Disease Identifier platform, which is designed to detect contagious tree diseases in forests and agricultural lands using machine learning, environmental sensors, and IoT connectivity. If diseases are detected, the platform will send out the results via text message.

The Tree Disease Identifier is designed around a LattePanda 3 Delta with a Wio Terminal and the SenseCAP K1100 kit (with vision AI module), which takes advantage of Edge Impulse's FOMO object detection model and the Twilio API that sends real-time alerts. The platform also has several environmental sensors that keep tabs on tree health, incluidng a Grove SCD30 (CO2, temperature, and humidity), a Grove SGP30 (tVOC and eCO2 levels) and a soil moisture sensor.

All of the hardware is packed within a 3D printed enclosure outfitted with a 7-inch display for visualizing captured images, which Edge Impulse's FOMO algorithm uses to identify diseases.

AI-Powered Trail Camera

Fifth guest speaker Daniel Legut, introduced his AI-powered trail camera that uses satellites and AI for animal research. The camera is an ideal solution for studying animals in remote locations with poor cellular reception, especially those animals afflicted with diseases such as Chronic Wasting Disease (CWD). This would allow rangers and farmers to keep tabs on how the disease spreads and take precautions to prevent it from spreading to healthy animals.

Legut designed his trail camera around the Orange Pi Zero 2W and takes advantage of a PIR sensor that triggers a web camera when motion is detected. Collected metadata, including perceptual difference values known as Hamming distances, is sent over satellite via a Starnote notecard through Notehub and processed by a Django web app.

When a significant number of images are stored, the camera is manually or community-moved to ridgelines or other areas where a cell signal is available, allowing image uploads. If an image's Hamming distance is above a threshold, a download request is triggered. The camera then resizes the original 480x640 PNG image to a 120x160 JPEG, converts it to base64, and transmits it in chunks over cellular networks. Once received, the image is reassembled on Django.

Wildlife Sanctuary Monitor

The sixth guest on the AI in the Wild webinar, Hendra Kusumah, demonstrated his Wildlife Sanctuary Monitor, which taps AI to monitor and maintain the sustainability of the sanctuary. His monitor takes advantage of Seeed Studio's SenseCAP K1100 kit and integrates sound recognition, visual object detection, environmental sensing, and LoRa-based wireless data transmission.

The monitor uses a Wio Terminal to coordinate various sensing modules and acts as a LoRa data gateway. For sound classification, Kusumah utilized the Wio Terminal's microphone sensor, along with a custom model he designed using Edge Impulse, which was trained to recognize the sounds of orangutans, rhinos, gunshots, and wildfires. The model uses MFE for feature extraction and Keras as the learning block, and once trained, it can be run live on the Wio Terminal or tested with real-time audio.

Visual detection is handled by a Grove AI Vision module, which supports onboard image classification and object detection. For wildfire detection, the monitor uses a Grove VOC and eCO2 Gas Sensor (SGP30) and a Grove Temperature and Humidity Sensor (SHT40), which is processed using an XIAO RP2040 microcontroller. The entire setup is powered by solar panels and an 18650 battery, making it easy to deploy in remote areas.

The Internet of Birds

The final guest for the AI in the Wild Impact Spotlights, Saudin Dizdarevic, walked us through his Internet of Birds project that uses AI to identify specific birds, including Bluejays, Cardinals and Titmice. Dizdarevic designed his platform using Seeed Studio's Grove Vision AI Module V2, which comes equipped with a camera and is capable of running machine learning models, allowing it to perform object classification locally without needing additional computing resources or cloud connectivity.

Dizdarevic created his custom model using 102 sample pics categorized into six classes, including the three aforementioned targeted bird types, which was done via Edge Impulse Studio. His model managed to achieve an impressive 93.8% accuracy, making it an ideal bird classification solution.

Conclusion

From smart lakes and sanctuary monitors to AI-powered collars and trail cameras, the AI in the Wild stream highlighted how off-the-shelf hardware and accessible AI tools have been utilized to mitigate some important environmental challenges. These innovative builds show how field-deployable tech doesn't require massive funds or data centers; rather, it just takes a clever mind and the right tools. As ecosystems continue to face growing threats, AI and IoT devices are proving to be invaluable tools that can provide real-world solutions to help safeguard our environments.