Automated Crop Dataset Clustering and Anomaly Detection Workflow

This n8n workflow is designed to set up and analyze crop datasets for anomaly detection using advanced clustering and embedding techniques. The process begins with a manual trigger, which allows users to test the workflow on demand. Key nodes include HTTP requests to interact with Qdrant cloud for managing collection points and their attributes, ensuring the dataset is accurately represented.

The workflow calculates the total number of points in the dataset and builds a distance matrix to evaluate the similarities between crop data points. It then converts this matrix into a sparse format to identify the most representative medoids within each crop cluster, using scoring mechanisms to determine the central points (medoids).

Additionally, the workflow integrates crop descriptions—both textual and visual—that are embedded via Voyage AI’s multimodal model to find the most similar crop image to each description, further refining cluster centers.

The process includes setting thresholds for anomaly detection based on dissimilar points within each crop cluster, enabling the system to identify outliers or anomalies effectively. Nodes for updating the dataset with threshold scores and medoid identifiers complete the setup.

This workflow is ideal for agricultural AI applications, where accurate crop classification, cluster analysis, and anomaly detection are vital for crop health monitoring, disease detection, and yield prediction, particularly when working with large image datasets and detailed crop descriptions.