ScannerVision: Scanner-based image acquisition of medically important arthropods for the development of computer...

Area of Science:

• Entomology
• Computer Vision
• Machine Learning

Background:

• Computer vision offers potential for rapid arthropod identification, but challenges exist in imaging large specimen numbers with sufficient quality for machine learning.
• Conventional methods using stereomicroscopes or macro lenses have narrow fields of view, hindering high-throughput processing.
• High image quality is crucial for training effective machine learning models for arthropod identification.

Purpose of the Study:

• To present a high-throughput, scanner-based method for capturing high-quality images of arthropods.
• To generate large, standardized datasets suitable for training machine learning algorithms.
• To demonstrate the method's effectiveness across various arthropod sampling techniques.

Main Methods:

• Utilized a charge-coupled device (CCD) flatbed scanner with optimized settings for arthropod imaging.
• Developed strategies for placing diverse arthropod samples onto the scanner bed.
• Captured high-resolution images balancing processing time and morphological detail.

Main Results:

• The scanner-based method successfully imaged arthropods from various collection methods (sticky traps, CDC light traps, BGS traps).
• Achieved morphological detail and resolution comparable to a stereomicroscope.
• Deep learning models trained on scanned images performed comparably to those trained on stereomicroscope images.

Conclusions:

• The scanner-based method provides a viable, high-throughput alternative for digitizing arthropod specimens.
• This approach generates high-quality datasets suitable for training machine learning identification models.
• The method's performance is comparable to traditional stereomicroscope imaging, facilitating large-scale biodiversity monitoring.