Automatic ladybird beetle detection using deep-learning models.

Area of Science:

• Ecology and agricultural entomology
• Computer vision and machine learning

Background:

• Invasive ladybird beetles pose significant threats to biodiversity and agriculture.
• Accurate and rapid identification is crucial for effective management strategies.

Purpose of the Study:

• To develop a two-step automatic detector for ladybird beetles in images.
• To establish the initial stage of an automated ladybird beetle classification system.

Main Methods:

• Image processing techniques including saliency maps, superpixel segmentation, and active contours were used to generate potential ladybird beetle locations.
• A deep convolutional neural network classifier was employed to confirm ladybird beetle presence within generated bounding boxes.

Main Results:

• The detector achieved 92% accuracy in bounding box generation and classification.
• An area under the receiver operating characteristic curve of 0.977 was obtained, demonstrating high performance.

Conclusions:

• The proposed automatic detector is a valuable tool for ladybird beetle detection.
• This system supports specialists in managing invasive ladybird beetle populations effectively.