Crop Anomaly Detection Workflow using Qdrant and Voyage AI

This n8n workflow is designed to detect anomalies in crop images by leveraging a combination of image embedding, clustering, and similarity comparison techniques. It primarily aims to identify crop images that do not conform to known classes, which can indicate potential anomalies or new crop types. The workflow integrates with Voyage AI for image embedding, Qdrant for storing and querying crop data, and triggers via image URLs to automate anomaly detection.

The process begins with an execution trigger that takes an image URL as input. The image is then embedded using Voyage AI’s multimodal embedding API, converting the visual data into a vector form. Next, the workflow queries the Qdrant vector database to find the most similar crop images based on distance scores, specifically comparing the embeddings to pre-established medoid cluster centers. The similarity scores are analyzed with a Python code node to decide whether the image matches an existing crop class or is potentially anomalous.

Throughout the process, various ‘sticky notes’ provide detailed explanations of the logic, such as the role of medoids and threshold scores, and the importance of these parameters for accurate anomaly detection. Additional steps include counting total points and crop class distribution within the dataset, which aids in understanding the dataset’s structure and the clustering setup.

This workflow is practically useful for agricultural researchers, crop quality control, or automated farming systems. It allows real-time validation of crop images uploaded via URL, facilitating quick detection of irregularities, disease, or new crop varieties, thus supporting data-driven decisions for crop management and research.