Recognition of agricultural objects by shape.

Area of Science:

• Computer Vision
• Image Processing
• Agricultural Security

Background:

• Agricultural contraband poses a significant threat to biosecurity.
• Current X-ray screening methods struggle to reliably detect diverse agricultural items.
• Distinguishing contraband from benign materials in luggage is challenging.

Purpose of the Study:

• To develop an algorithm for enhancing the outlines of agricultural contraband in X-ray images.
• To enable automated detection of items with elliptical cross-sections.
• To improve the accuracy of contraband detection in passenger baggage.

Main Methods:

• Proposed an algorithm utilizing the erosion of the absolute gradient for cross-section recognition.
• Employed local convolution calculations for efficient processing.
• Tested the algorithm on computed and real-world X-ray images, including obscured items.

Main Results:

• The algorithm successfully enhances outlines of agricultural contraband, even for small items (1-2 cm).
• Detection is effective despite image noise, varying object sizes, orientations, and obscuration.
• The method distinguishes contraband with elliptical cross-sections from other materials.

Conclusions:

• The proposed algorithm offers a robust method for enhancing agricultural contraband outlines in X-ray images.
• This technique shows promise for improving automated detection systems in security screening.
• The algorithm's effectiveness under realistic conditions supports its practical application.