改进深度置信网络的苹果内部品质评价

目的：解决苹果近红外光谱存在大量冗余信息和苹果内部品质评价精度较低的问题，提高苹果内部品质评价的精度。方法：提出一种连续投影法的特征波长筛选与灰狼优化算法改进深度置信网络(GWO-DBN)的苹果内部品质评价模型。针对苹果光谱数据具有维度高而复杂的特点，分别对比全波段和主成分分析法、连续投影法等筛选特征波长的结果，确定苹果光谱特征波长筛选方法；针对深度置信网络(DBN)模型性能受参数设定的影响，运用灰狼优化算法(GWO)对DBN模型参数进行优化选择，提出一种连续投影法的特征波长筛选与GWO-DBN的苹果内部品质评价模型。结果：与中粒子群算法改进深度置信网络(PSO-DBN)、遗传算法改进深度置信网络(GA-DBN)和DBN相比，基于GWO-DBN的苹果内部品质评价的准确度最高。结论：GWO-DBN算法可以有效提高苹果内部品质评价的准确率。

Evaluation of apple inner quality based on improved deep belief network

Objective:In order to resolve a lot of redundant information and low precision of apple internal quality evaluation existed in apple near infrared spectroscopy, improving the precision of apple internal quality evaluation.Methods:A new apple inner quality evaluation model based on deep belief network (DBN) and grey wolf optimization algorithm was proposed. According to the characteristic of high dimension and complexity of apple spectral data, the method of selecting characteristic wavelengths of apple spectral data was determined by comparing the results of selecting characteristic wavelengths of full-band, principal component analysis and continuous projection. The parameters of DBN model were optimized by GWO Algorithm, and a continuous projection method for feature wavelength selection and GWO-DBN model for apple inner quality evaluation were proposed.Results:Compared with PSO-DBN, GA-DBN and DBN, the accuracy of apple inner quality evaluation based on GWO-DBN was the highest.Conclusion:This algorithm can effectively improve the accuracy of apple inner quality evaluation and provide a new method for apple inner quality evaluation.