| Summary: | 碩士 === 慈濟大學 === 醫學資訊學系碩士班 === 107 === Image recognition and identification is one of the most popular areas in the broad field of imaging sciences. There are many image identification systems for plant and animal identification. Most plant identification systems are only classified into families and genera, however the identification for species is relatively unclear. In the application of biology, it is very practical to explore new species when the image identification system can classify species. In this study we use the deep learning techniques, convolutional neural networks, to build an agave species classification system. We collected ten varieties of agave images as training materials. The training data include the whole plant images (about 20,000 images) and leaf images (about 12,000 images). We divided these data into 5 datasets. Some dataset are original images and some dataset are modified by cutting the background patterns. The models of convolutional neural networks are trained by these datasets individually. Finally, we compare these different models and discussed the accuracy rates.
|
|---|
