Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
The recognition of Gastrodia elata variants based on machine vision
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2021 ASABE Annual International Virtual Meeting 2100345.(doi:10.13031/aim.202100345)
Authors: Zimei Zhang, Zhian Zheng, Andi Chen, Shanyu Wang, Yiyao Xu
Keywords: Different variants; Features extraction; Gastrodia elata; Image recognition; Support vector machine.
Abstract. Recognition of Gastrodia elata Blume (G. elata) variants is a very important aspect in evaluating its quality. However, the traditional evaluation method based on sensory is too subjective. Machine vision technology can identify images of G. elata objectively and efficiently. Therefore, this study proposed a recognition method of G. elata based on machine vision, which can successfully achieve features extraction and image recognition of three common G. elata variants which included red G. elata, black G. elata and hybrid G. elata. Firstly, 435 images of G. elata were captured. After preprocessing of denoising and segmentation, color features (RGB color moments and HSV color moments), shape features (length, width, width-length ratio, area and 7 invariant moments) and texture features (energy mean, energy standard deviation, entropy mean, entropy standard deviation, mean moment of inertia, the standard deviation of the moment of inertia, correlation mean and correlation standard deviation) were obtained. These three kinds of single features were combined into four kinds of compound features, namely, color-texture features, color-shape features, texture-shape features, and color-texture-shape features. Then, through the establishment of thirty kinds of discrimination model including decision trees, Naive Bayes and Support Vector Machine and others, the samples were trained and tested based on 3 kinds of single features and 4 kinds of compound features. The ratio of the number of training samples to the number of test samples was 3:1 and 10-fold cross validation was applied to training model. Next, the discrimination model with the highest accuracy for each features was selected. Finally, the optimal model for each features was tested. The results showed that the optimized support vector machine model based on color-texture-shape features has the best effect and the accuracy rate was as high as 98.1%. This research effectively identified three common G. elata variants, provided a theoretical basis for better application of machine vision technology to evaluate the quality of G. elata, and created a new possibility to monitor G. elata quality more conveniently.
(Download PDF) (Export to EndNotes)