Data-driven Solutions for Driving Away Algae

Many cities in the southern reaches of the Yangtze River are home to vast water networks dotted with scenic lakes and tourist-targeted lakeside restaurants and entertainment. The boost to city incomes is accompanied by blooms of blue-green algae in spring, summer, and fall, which degrade aquatic ecology and deter tourists with the unpleasant look and smell.

In the past, algae management relied on manual observation, manually operated unmanned aerial vehicle (UAV) inspections, and aquatic weed harvesters. However, as algae floats with the wind, these method aren’t sufficient.

The digital foundation of Huawei’s smart city approach can be applied to local ecology through AI-enhanced UAVs. Invoking and orchestrating UAV fleet management and route planning capabilities makes regular, task-driven algae inspections possible.

Visible light imaging 

Examining the difference between the optical properties of a blue-green algal bloom and unaffected water shows that the refraction of algae improves significantly in infrared and near-infrared bands, but remote sensing reflectance (Rrs) is higher in the green band, as algae contains chlorophyll-a. The most common way to tell algae and normal water apart is to reconstruct simple features, such as band variances, based on the data of different bands and multispectral imaging. Differences can be identified using classical imaging segmentation methods or classification methods, such as support vector machine (SVM), for machine learning.

UAV-mounted cameras normally use visible-light images, which requires machine vision. Unlike multispectral images, visible-light images don’t cover infrared bands and the green band is basically indistinguishable, making it difficult to accurately identify blue-green algae. As UAVs are deployed outdoors, their detection functionality is also affected by factors like shooting angle, distance, and lighting conditions. The color of some algal blooms can appear almost identical to the surrounding water due to strong light reflection, so sample images don’t accurately show color variance. City managers must frequently change the cameras in the UAVs or mount multiple cameras on a single UAV, which boosts identification efficacy but reduces efficiency and is limited by the battery life of UAVs.

To address this problem, we must identify the optimal technical path for visible-light-based visual recognition. Huawei's Intelligent Operation Center (IOC) features big data and AI capabilities that use a deep neural network algorithm and multi-scale fusion network to extract datasets of algal images from various types of terrain, ranging from simple water-surface and shoreline to more complex water flows through median strips or shoreline neighborhoods. Complex models plus constant learning and training allow Huawei's IOC to more accurately and efficiently detect algae.

Figure 1: Algae floating on a lake surface (simple identification) 

Figure 2: Algae floating on a narrow waterway surface (more complex identification) 

Figure 3: Shoreline algae

Figure 4: Algae on a neighborhood river surface identified by AI-powered UAVs

Figure 5: Variations in bio-optical properties (Rrs) of turbid water, in-water algae, and floating bloom

Powered by IOC for algae prevention

The IOC enables municipal authorities to deploy large-scale intelligent services. Its AI platform can centrally develop algorithm models and provide full-lifecycle management services. The IOC integrates machine vision identification capabilities into various APIs that can be orchestrated and coordinated for the reuse, combination, and large-scale development of user applications. The solution minimizes manual intervention and O&M costs, and supports integrated operations. 

Key technologies

Video data collection: The video cloud supports H.264 and H.265 video coding formats, receives and stores real-time video streams from UAVs over the Internet, and exposes video capture and retrieval services to the AI platform through APIs.

5G connection: Powered by 5G's low latency and high bandwidth, UAVs support 4K ultra-HD image transmission and AI-enabled real-time identification and alerts.

AI algorithm scheduling: The AI platform can invoke the video stream data sent from the UAVs to calculate the longitude, latitude, severity, and distribution of algal blooms based on video timestamps and flight data. It calculates the accumulated algae area discovered by the UAVs, captures algae images and video clips, and sends different levels of alarms to the work order preprocessing and distribution system.

Application integration and message integration: The API gateway acts as the service platform that receives algae identification data sent from the AI platform. It packages other service logic data, such as real-time UAV location, and uses the messaging middleware service to place the data in message queues. It then sends the message queues to visualized user interfaces or other service systems that subscribe to the message service.

Big data analytics and processing: The deep learning module underpins algae identification algorithms. Based on data governance, it packs algae data into different subjects and defines data dimension structures based on different service requirements to provide customized content.

UAV fleet management capability: The UAV hangar and management software provided by Huawei's partners enables fully autonomous UAV scheduling and driving. The hangar supports automatic take-off, landing, and charging. Flight packets can be sent back to the IOC in real time to open up the scheduling capability.

Optimized flight: The IOC controls the hangar and sets fleet flight routes and mission plans based on factors such as key areas, time, and tasks. The UAVs take off automatically and the AI algorithm automatically calculates the area and location of algae and algal protein concentration. Automated incident reports and work orders enable prompt, accurate, and efficient harvesting. The IOC normalizes the process of sending information such as algae alarms and work order closure statistics to different users in the form of SMS messages or printed reports.

One city in Jiangsu deployed an IOC in 2020. It covers 19 square kilometers of water in the city with two UAVs, averaging 12 flights per day. Without increasing staffing or vessels, 36,000 tons of blue-green algae have been harvested since the system was deployed and 15.1 tons of algae sludge were divided, boosting efficiency by more than 45%. 

Stopping water pollution at the source

Seasonal changes and local weather patterns can affect blue-green algae growth, but it’s the industrial wastewater and domestic sewage that cause the major spikes. Wastewater, for example, can lead to excessive nutrients such as nitrogen and phosphorus, encouraging year-round blooms in river basins with concentrations of businesses and restaurants. Harvesting teams could clear the areas daily, but the blooms would return the next day. So, the solution is to control the sources of pollution. 

To get the most out of technology, water needs to be able to purify itself through restored aquatic food chains. We must control industrial wastewater and domestic sewage discharge, and treat wastewater to remove nutrients that cause blue-green algae spikes. The IOC’s big data platform performs vertical and horizontal data mining, with blue-green algae alarms and clean-up tasks generated based on wastewater discharge standards and data for sewer monitoring and enterprise violators that’s provided by various authorities. The IOC can profile enterprises located along riverways to enable wastewater governance from the source.

For data quantification and modeling, water pollution models can be created based on the data, such as algal protein density, pH, electrolyte, chlorophyll, and turbidity, obtained by IoT sensors on unmanned surface vessels (USVs) during river cruises. This data can train and optimize an AI algae visual identification model, and identify enterprises that discharge wastewater, components in wastewater, discharge locations, and the raw materials and processes they use. The data can be used to study how these factors relate to the distribution, area, and concentration of seasonal algae outbreaks and the number of harvesting operations, and can help identify illegal discharge through concealed pipes or leaks. The process requires the traversal of various combinations of possibilities, simulation-based inference, and decision-making.

USVs can send up-close monitoring data back to the IOC through the 5G network. UAVs can automatically respond to the requests of USVs and follow their tracks in the sky. In-depth detection and wide-area monitoring by both UAVs and USVs can make water management more time-efficient and intelligent.

Thermal mapping services provided through high-altitude remote sensing satellites can be used to periodically analyze the distribution of algae on the surface of large bodies of water. Along with weather data, this enables targeted trend analyses, mission planning and route design for cruise flights, and harvesting operations, which minimize the time and labor required for these operations. This space-sky-surface collaboration not only applies to blue-green algae identification, but also to urban water resource management, ecological restoration, environmental protection, disaster prevention, and more.