This n8n workflow is designed for automated crop anomaly detection based on image analysis. It processes an input image URL, creates an embedding vector using Voyage.ai’s API, and compares it to a pre-existing dataset of crop images stored in Qdrant vector database. The workflow determines whether the image belongs to a known crop class or is an anomaly, providing a straightforward decision message.

The workflow begins with an “Execute Workflow Trigger” node, which receives an image URL as input. This URL is passed to the “Embed image” node, which sends the image for embedding via Voyage’s API to generate a high-dimensional vector representation of the image.

Next, the workflow retrieves key dataset information, such as the total number of crop classes and their representative medoid vectors, from Qdrant. It calculates similarity scores between the new image embedding and the cluster centers (medoids) stored in Qdrant through the “Get similarity of medoids” node, which queries the Qdrant API.

The similarity scores are then processed in a Python code node, “Compare scores,” which identifies the closest crop class based on threshold scores. If the new image’s similarity exceeds the pre-set thresholds, it is assigned to that crop class. If not, the system flags it as a potential anomaly.

Additional notes include detailed instructions on dataset setup involving Google Cloud Storage and Qdrant cloud configurations, emphasizing that this pipeline is adaptable to various image datasets. The workflow is useful for agricultural monitoring, quality control, and research, enabling automatic, real-time identification of anomalies in crop images and potentially detecting new or unexpected crop varieties.