[3/3] Anomaly Detection Tool (Crops Dataset)

Workflow Name

[3/3] Anomaly Detection Tool (Crops Dataset)

Key Features and Highlights

This workflow is an automated tool designed for anomaly detection in crop images. By inputting the URL of any crop image, it leverages a multimodal embedding model to generate image vectors, which are then compared against a pre-built crop image database stored in the Qdrant vector database. This process determines whether the image belongs to a known crop category or represents an anomaly (unrecognized crop).

• Supports a wide range of crop categories, including pearl millet, tobacco, cherry, cotton, banana, cucumber, corn, wheat, clove, buckwheat, olive tree, soybean, coffee, rice, lemon, mustard oil, mung bean, coconut, chickpea, pineapple, sugarcane, sunflower, chili, lotus seed, jute, papaya, tea, cardamom, almond, and more.
• Integrates Voyage AI’s multimodal embedding API with Qdrant’s cloud vector database to enable efficient feature extraction and similarity search of images.
• Employs preset threshold-based decision rules to accurately identify anomalous crop images, enhancing crop quality monitoring and anomaly alert capabilities.

Core Problem Addressed

Traditional crop image classification methods struggle to effectively identify unknown or anomalous varieties. This workflow solves the challenge of automatic anomaly detection within crop image datasets by using vector retrieval and threshold comparison techniques. It helps users quickly pinpoint potential new varieties or abnormal cases, preventing data confusion and misclassification.

Application Scenarios

• Agricultural Monitoring and Quality Control: Automatically detect and identify anomalous crops to improve product quality assurance.
• Crop Variety Identification and Management: Assist researchers and agronomists in rapid classification and recognition of crop images.
• Agricultural Data Analysis and Automation: Combine with cloud services to automate anomaly detection in crop images, boosting operational efficiency.
• Image Dataset Maintenance: Timely discover and remove or label anomalous images in datasets to maintain data integrity.

Main Workflow Steps

1. Trigger Workflow — Receive image URL input via the Execute Workflow Trigger node.
2. Set Variables — Configure parameters for accessing the Qdrant cloud vector database (e.g., URL, collection name, threshold values).
3. Retrieve Database Info — Query Qdrant for the number of crop image categories and total data points.
4. Generate Image Embeddings — Call Voyage AI’s multimodal embedding API to convert the input image into a vector representation.
5. Similarity Search — Use the generated vector to perform a similarity search in the Qdrant database, obtaining similarity scores against category medoids.
6. Score Comparison and Decision — Use a Python code node to compare similarity scores with the threshold; if all scores fall below the threshold, classify the image as anomalous; otherwise, return the closest matching crop category.
7. Output Results — Return a text message informing the user whether the image is anomalous and its best matching category.

Involved Systems and Services

• Qdrant Cloud: Vector database for storing and retrieving crop image vectors and category medoids.
• Voyage AI Embeddings API: Multimodal embedding service for converting images into vector representations.
• n8n Automation Platform: Connects and orchestrates nodes to automate the entire workflow.

Target Users and Value

• Agricultural researchers and technicians, supporting crop variety identification and anomaly monitoring.
• Data scientists and machine learning engineers, facilitating anomaly detection and data cleaning in image datasets.
• Agricultural production managers, enabling automated crop quality inspection and risk alerting.
• Automation developers, providing a rapid solution to build vector retrieval-based anomaly detection systems.

---

Built upon advanced vector retrieval technology and multimodal embedding models combined with cloud data storage, this workflow delivers efficient anomaly detection for crop images. It offers excellent scalability and applicability, providing robust technical support for intelligent management in the agricultural sector.