A novel transfer method with neural network architecture searching to predict asymptomatic of akizuki pear cork spot disorder on near-infrared spectroscopy

Akizuki pear cork spot disorder is a physiological disease, the ability to effectively diagnose the diseased fruit can directly affect the fruit quality. To improve the diagnostic performance of asymptomatic samples, this paper proposes a novel spectral transfer modelling method based on neural network architecture searching, named TranNAS_NIR. The experiments show that the model by the proposed TranNAS_NIR method is more effective. The accuracy of the target domain test set reaches 82.61%, which is 5.3% better than the model with transfer component analysis (TCA) and 10.5% better than the model constructed only relying on the target domain data and without the transfer learning method. The novel transfer method with neural network architecture search proposed in this paper for the asymptomatic prediction model of Akizuki pear cork spot disorder also further provides a theoretical basis for the NIR model improvement of other asymptomatic diseases.

Automated summary

This paper proposes a novel spectral transfer modelling method based on neural network architecture searching for the asymptomatic prediction of Akizuki pear cork spot disorder. The experiments show that the model by this method is more effective, achieving an accuracy of 82.61%.