Description
Overview
This KNN classifier workflow provides a precise automation workflow for satellite image classification based on k-nearest neighbors (KNN) analysis. This orchestration pipeline processes an input image URL through vector embeddings and similarity search to classify land-use types from a predefined dataset, triggered by an Execute Workflow Trigger node.
Key Benefits
- Enables automated land-use image classification using vector embedding similarity search.
- Employs a majority voting mechanism to ensure robust class determination in the orchestration pipeline.
- Resolves classification ambiguities via an iterative increase in nearest neighbor queries.
- Integrates with Qdrant vector database for scalable and efficient nearest neighbor retrieval.
Product Overview
This KNN classifier automation workflow initiates with the receipt of a satellite image URL via the Execute Workflow Trigger node. The image URL is then forwarded to the Voyage.ai Multimodal Embeddings API, which generates a high-dimensional vector embedding representing the image’s content using the “voyage-multimodal-3” model. The workflow sets up query parameters for Qdrant, including the embedding vector, collection name (“land-use”), and initial number of neighbors (limitKNN = 10). Subsequently, it queries the Qdrant cloud vector database to retrieve the closest matching images with their associated land-use labels. A Python-based majority vote node processes these neighbors to identify the most frequent classes. If a tie occurs between the top two classes and fewer than 100 neighbors have been queried, the workflow increments the neighbor count by 5 and repeats the query loop until a clear majority is reached or the limit is met. The final classification is extracted and returned synchronously to the calling process. Error handling leverages native n8n platform defaults without custom retry logic. Authorization to external APIs is managed via HTTP header and predefined credential types, ensuring secure integration without persistent data storage.
Features and Outcomes
Core Automation
The KNN classifier orchestration pipeline accepts an image URL input, computes its embedding, and deterministically identifies the land-use class based on nearest neighbor voting. The decision logic includes a tie-breaking loop that expands neighbor retrieval until resolution.
- Single-pass embedding generation via Voyage.ai API ensures consistent vector representation.
- Majority Vote node applies frequency counting on neighbor labels using Python code.
- Iterative logic with dynamic neighbor limit increment prevents ambiguous classifications.
Integrations and Intake
The no-code integration pipeline interfaces with Voyage.ai for embedding extraction and Qdrant Cloud for vector similarity search. Authentication uses HTTP header-based API keys and predefined Qdrant credentials. Input requires a valid satellite image URL field named “imageURL”.
- Voyage.ai Multimodal Embeddings API for image vectorization.
- Qdrant Cloud vector search API for nearest neighbor retrieval.
- n8n Execute Workflow Trigger node for JSON input intake with image URL.
Outputs and Consumption
Outputs are provided synchronously as a JSON object containing the classified land-use label under the “class” key. The workflow returns a single definitive classification per input image after tie resolution or query limit reached.
- JSON response includes “class” key with the identified land-use category.
- Classification result derived from majority voting of nearest neighbor labels.
- Synchronous response suitable for direct workflow chaining or API consumption.
Workflow — End-to-End Execution
Step 1: Trigger
The workflow is initiated via the Execute Workflow Trigger node, which accepts a JSON payload containing an “imageURL” field. This URL points to the satellite image to be classified.
Step 2: Processing
The Image Test URL node extracts the input URL for embedding. The Embed image node sends a POST request to Voyage.ai’s embedding API with the image URL, receiving a high-dimensional vector embedding in JSON format. Basic presence checks ensure the required “imageURL” is defined before proceeding.
Step 3: Analysis
Qdrant is queried with the embedding vector and an initial limit of 10 neighbors. The Majority Vote node executes Python code to count the two most frequent land-use labels among neighbors. The Check tie node evaluates if the highest two counts are equal and if the neighbor count is ≤ 100, triggering an incremental loop to increase neighbor retrieval by 5 until a decisive majority or the limit is reached.
Step 4: Delivery
Upon resolving any ties or reaching the maximum neighbor count, the Return class node extracts the top label and outputs it as the final classification. The response is synchronous, delivered back to the calling workflow or API consumer as a JSON object with the classification result.
Use Cases
Scenario 1
Organizations need to classify satellite imagery for land-use monitoring. This automation workflow ingests image URLs, computes embeddings, and classifies images among predefined categories, enabling structured land-use data generation in a single synchronous response cycle.
Scenario 2
Environmental analysts require consistent classification of satellite images without manual review. By querying a vector database of labeled images, the workflow determines the most likely land-use type through majority voting, reducing manual processing and improving classification reliability.
Scenario 3
Developers integrating image classification into geospatial platforms can leverage this no-code integration to embed images and query similar labeled data with automated tie resolution, ensuring deterministic and reproducible classification outputs for diverse satellite datasets.
How to use
To deploy this KNN classifier workflow, import it into an n8n instance and configure credentials for Voyage.ai API and Qdrant Cloud. Provide image URLs in JSON format to the Execute Workflow Trigger node. Set the initial limitKNN parameter as needed. Run the workflow; it processes the input image, queries nearest neighbors, performs majority voting, and outputs the classified land-use label. Expect synchronous JSON responses containing a “class” field indicating the classification.
Comparison — Manual Process vs. Automation Workflow
| Attribute | Manual/Alternative | This Workflow |
|---|---|---|
| Steps required | Multiple manual image reviews and cross-referencing labels. | Single automated classification cycle with iterative neighbor queries. |
| Consistency | Subject to human error and varying criteria. | Deterministic majority voting with dynamic tie-breaking logic. |
| Scalability | Limited by manual capacity and human resources. | Scales with vector database and API throughput. |
| Maintenance | Requires ongoing manual dataset updates and training. | Automated embedding and query updates using external APIs and Qdrant. |
Technical Specifications
| Environment | n8n automation platform with external API access |
|---|---|
| Tools / APIs | Voyage.ai Multimodal Embeddings API, Qdrant Cloud vector search |
| Execution Model | Synchronous request-response with iterative querying |
| Input Formats | JSON with “imageURL” string field |
| Output Formats | JSON with classified “class” string field |
| Data Handling | Transient image embedding and neighbor payloads; no persistent storage |
| Known Constraints | Max neighbor queries capped at 100 to resolve classification ties |
| Credentials | HTTP header API key for Voyage.ai; predefined API key for Qdrant Cloud |
Implementation Requirements
- Valid API credentials for Voyage.ai embedding service and Qdrant Cloud vector database.
- Accessible satellite image URLs provided in the expected JSON input format.
- Network access allowing outbound HTTPS requests to Voyage.ai and Qdrant endpoints.
Configuration & Validation
- Configure HTTP Header Authentication credentials for the Voyage.ai API node.
- Set up Qdrant API credentials and verify access to the “land-use” collection.
- Test workflow execution with sample JSON input containing valid “imageURL” fields to confirm classification output.
Data Provenance
- Triggered by Execute Workflow Trigger node accepting JSON with “imageURL”.
- Embedding generated by Embed image node via Voyage.ai API using the “voyage-multimodal-3” model.
- Nearest neighbor retrieval conducted by Query Qdrant node against “land-use” collection on Qdrant Cloud.
FAQ
How is the KNN classifier automation workflow triggered?
The workflow is triggered by receiving a JSON payload with an “imageURL” field at the Execute Workflow Trigger node.
Which tools or models does the orchestration pipeline use?
The pipeline uses the Voyage.ai Multimodal Embeddings API with the “voyage-multimodal-3” model and the Qdrant Cloud vector database for nearest neighbor searches.
What does the response look like for client consumption?
The workflow returns a synchronous JSON response containing a “class” key with the determined land-use classification label.
Is any data persisted by the workflow?
No data is persisted by the workflow; embeddings and query results are transient and processed in-memory during execution.
How are errors handled in this integration flow?
Error handling relies on default n8n platform behavior; no custom retry or backoff logic is implemented within the workflow.
Conclusion
This KNN classifier automation workflow provides a deterministic method to classify satellite images into predefined land-use categories by leveraging vector embeddings and similarity search. Its iterative neighbor expansion handles classification ambiguities efficiently, delivering a single synchronous output per input image. This workflow depends on external API availability for embedding generation and vector querying, which is a necessary operational constraint. Overall, it enables structured, reproducible land-use classification without manual intervention, supporting scalable geospatial data workflows.
Reviews
There are no reviews yet.