| Summary: | 碩士 === 國立嘉義大學 === 生物機電工程學系研究所 === 106 === Abstract
This study aims to mainly use image processing to complete honeycomb classification. Automatic detection of the distribution of honeybee larvae, honey and pollen and their location helps to properly manage groups of bees, remove larvae and automatically collect honey for resolving problems posed by an aging workforce of beekeeping and increasing production quantity and quality of bee products in order to increase the income of beekeepers. This study first used support vector machine (SVM) based on HSV color model to classify honeycomb images into 8 pixel types: larvae, egg, honeycomb wall, honeycomb base, sealing cover, honey, reflective light and pollen. The honeycomb wall acquired from the pixel classification process was not complete in general, so we failed to completely divide them into separate honeycombs. In order to resolve this issue, a 4×7 hexagonal grid that simulates a honeycomb wall was compiled as the wall template to contrast with an incomplete honeycomb wall for matching algorithms to identify the two best matching positions. This wall template replaced the original honeycomb wall for independent separation of each honeycomb cell. Finally, the proportion of eight pixel types and six color feature parameters were acquired from each honeycomb cell and SVM was applied to classify the honeycombs into six types: larvae, egg, sealing cover, empty honeycomb, honey and pollen. The experimental results indicated that the recognition of six honeycombs reached 95.46%, 40.71%, 86.32%, 64.37%, 87.01%, and 83.54%, respectively.
Keyword: image processing, honeycomb, support vector machine, image matching
|
|---|
