Identifying Irregular Potatoes Using Hausdorff Distance and Intersection over Union.

Area of Science:

• Computer Vision
• Agricultural Technology
• Image Processing

Background:

• Irregularly shaped potatoes limit further processing and value addition in the agricultural industry.
• Existing methods for identifying potato shape may be complex or computationally intensive.

Purpose of the Study:

• To develop and evaluate an efficient method for identifying irregular potatoes using image analysis.
• To compare the effectiveness of different shape characterization metrics for irregular potato detection.

Main Methods:

• Potato images undergo preprocessing including resizing, segmentation, and filtering to extract contour information.
• A least-squares fitting method is employed to fit an ellipse to the extracted potato contours.
• Shape similarity is quantified using perimeter ratio, area ratio, Hausdorff distance, and intersection over union (IoU).

Main Results:

• Hausdorff distance and IoU demonstrated superior characterization ability compared to perimeter and area ratios for identifying irregular potatoes.
• Using Hausdorff distance alone achieved high performance (Precision: 0.9423, Recall: 0.98, F1: 0.9608).
• Using IoU alone resulted in excellent recognition rates (Precision: 1, Recall: 0.96, F1: 0.9796).

Conclusions:

• The proposed ellipse-fitting method effectively identifies irregular potatoes using Hausdorff distance or IoU as single feature parameters.
• This approach simplifies feature complexity and reduces computational effort compared to high-dimensional methods.
• The algorithm's reliance on simple threshold segmentation eliminates the need for data training, saving execution time.