Research On Blueberry Internal Quality Detection Method Based On Hyperspectral Imaging Technology

In recent years, blueberries with its abundant nutritional value and health value has been favored by the consumers. With the improvement of material life, people gradually increased blueberries’quality requirements. But now the manual sorting, grading can only rely on the size of the blueberry, color, shape of screening. The blueberry’s internal quality can’t distinguish. In recent years, Hyperspectral Imaging Technology is a new rise subject, which is fuse the image technology and optical technology. Many research scholars at home and abroad applied it to the detection of fruit quality, put forward a lot of fruit nondestructive testing methods, and achieved good results. In this paper, with fresh blueberries as the research object, using Hyperspectral Imaging Technology on internal quality (sugar and hardness) of blueberries fast nondestructive testing. For hyperspectral implementation blueberry on-line detection provides technical and theoretical basis.In this paper, blueberry as research subjects which is the most representative among berries. First, using self-built hyperspectral imaging device to capture hyperspectral image of blueberry, measurement Blueberry Brix and hardness after collected blueberry’hyperspectral images. Secondly, corrected blueberry hyperspectral image, used software to select rectangular region of interest (the rectangular area of 30×30), calculated the average spectral reflectance for all points within the ROIs, removed the area affected by the larger noise after observed the average spectral reflectance curve, and select between 500nm-1000nm spectral data as a whole data. Thirdly, study seeked to simplify and optimization data dimensionality reduction method-Successive Projection Algorithm (SPA). In the prediction of brix, selected the 12 characteristic wavelengths, and reduced 96.92% of the amount of data. In the prediction of hardness, selected the 8 characteristic wavelengths, and reduced 97.95% of the amount of data. Finally, used BP neural network to build prediction models:brix、hardness prediction models based on full band-BP neural network, brix, hardness prediction models based on characteristic bands-BP neural network. And compared the model predictions.Research indicates, the models has all achieved good forecast results, compared to the full band-BP neural network prediction models, the characteristic bands-BP neural network prediction models getted better prediction. Used feature band extraction method to established blueberry Brix, hardness prediction models can not only reduced the dimensionality of the data, reduced the computational model, but also achieved good predictions.