This paper presents a practical recognition method based on deep learning techniques, for fungal species, through Scanning Electron Microscopy (SEM) images. A small number of images was acquired. To circumvent the issue of not having many samples, a method of generating the training set is proposed to increase target signatures and optimize the baseline quality of inputs for object recognition. To tackle the challenge of detecting varied scale targets, a sophisticated and powerful Convolutional Neural Network (CNN) based on faster region R-CNN, with the prepared training dataset, was trained. In this study, the datasets for five different species of Aspergillus were previously collected via SEM. The proposed method is applied to identify the spore structures – conidia – in the images so as to recognize the species respectively. The initial experimental results show that the developed method can qualitatively and quantitatively identify the relevant species effectively, being of major importance for the development of easier diagnostic and identification tools in mycology.
|Number of pages||6|
|Journal||International Journal of Pharma Medicine and Biological Sciences|
|Publication status||Published - 1 Jan 2022|
- Aspergillus conidia
- convolutional neural networkk
- object recognition
- scanning electron microscopy