| The freshness and edible quality of rice,as one of the reserve grains in China,are closely related to its economic value.Therefore,identifying the freshness of rice is crucial.Traditional methods for detecting rice freshness,such as conductivity method,electronic nose method,and hydrogen peroxide enzyme detection method,are cumbersome and have long detection cycles.They cannot meet the current demand for rapid quality testing of agricultural products.Therefore,an efficient and accurate method for detecting rice freshness is urgently needed.Terahertz(THz)technology,as a new type of detection technology,has been widely used in the quality inspection of agricultural products due to its advantages of fast detection analysis speed,high spectral resolution,and strong penetration ability.In this dissertation,a recognition method based on the fusion of spectral and image information is proposed to deal with problems such as low efficiency and accuracy in rice freshness detection.Establishing a recognition model for rice freshness,which provides technical support for rapid and accurate identification of rice freshness.The main research contents are as follows:(1)To address the problem of insufficient specific correlation research between rice freshness and fatty acids,a THz-based analysis method for rice freshness correlation is proposed.It is used to investigate the interrelationship between rice freshness and fatty acids.Three types of fresh rice and three types of fatty acid samples were collected using THz time-domain spectroscopy and THz reflection imaging technology.The moving window correlation coefficient method was used to study the correlation between fatty acids and rice freshness.The results show that there is a significant negative correlation between palmitic acid and rice freshness.(2)To address the problems of the VGG19 network in the spectrum and image recognition of rice with different freshness,such as low nonlinear data processing capability and insufficient multiscale feature extraction,an improved method based on the Inception-ResNet-V2 network is proposed.It is used to improve the accuracy of network recognition.The method uses VGG19 network as the basic network structure and introduces the Inception-ResNet-A module from the Inception-ResNet-V2 network after the first convolution block of VGG19.Experimental results show that the evaluation metrics of the improved network are significantly improved compared to those of the unimproved network.(3)To address the problems of limited detection capability of THz spectral and image analysis and low recognition accuracy,a feature fusion method based on the fusion of multiple information is proposed.It is used to construct a multi-information fusion model.Based on the spectral and image characteristics of different freshness rice samples in the THz band,a THz spectral and image information fusion model is established using multi-information fusion technology.The identification of rice freshness is achieved,and the results of various fusion models are compared to obtain the optimal freshness detection model.Experimental results show that the spectral and image feature fusion model performs better in detecting rice freshness than single-information models. |
