A multimodal quality inspection system based on 3D hyperspectral and X-ray imaging for onions

摘要

This paper reports a multimodal machine vision system developed for quality inspection of onions. The system integrates hyperspectral, color, 3D, and X-ray imaging technologies to evaluate quality factors of onions holistically and nondestructively. ALabVIEW program was developed to acquire color image, NIR spectral image, depth image, and X-ray images of onions, and measure the weight of onions. Three types of onions were tested in this study: healthy onions without inoculation, inoculated with Burkholderia cepacia, and inoculated with Pseudomonas viridiflava. Data fusion algorithms based on image processing, statistical, and machine learning techniques were applied to estimate the maximum diameter, volume, and density of onions, and detect onionswith defects. Results showed that the system accurately measured the diameter (RMSE=1.7 mm), volume (accuracy=96.9%), weight (RMSE=3.7 grams), and density (RMSE=0.03 gram/cm~3) of onions. Using image features selected from onion X-ray and spectral images, two support vector machines cascaded at the decision level successfully classified 81.58% healthy and defective onions. The classification tree utilizing features combined at the feature level distinguished 84.21% onions. Classifiers based on multisensor data fusion achieved much higher detection rates of defective onions than those of classifiers using single sensor/camera (78.95% and lower). The work demonstrated that the multisensor-based system can evaluate both external and internal quality parameters of onions, and provided a base for the further development of fully automated robotic system for onion quality inspection. The system and methods presented in this article are also potentially applicable to quality inspection of other agriculturalproducts.