摘要
针对温室环境数据的维度高、冗余性强,导致数据处理存在压缩比低和峰值信噪比较高的问题,提出基于自编码神经网络高阶特征提取的温室环境因子高维数据压缩方法。应用改进回归方程,填补温室环境因子数据中的缺失值,针对深度自编码神经网络的内部协变量迁移现象,加入自适应平衡层,结合小批量梯度下降法,构建深度自适应平衡自编码神经网络,提取温室环境因子高阶特征,基于矢量量化思想,判断相对误差,通过实施新码书计算,获得各划分的质心,根据码书训练结果,设计高维数据压缩方法。结果表明,当数据量超过50 GB时,所设计方法的压缩比下降0.7个百分点,降幅为3.8%,整体压缩性能表现优异;峰值信噪比随着采样率变大并未大幅下降,仅降低4 dB,降幅为7.5%,压缩峰值信噪比具备更优的重建保真度。该方法具有更高的压缩比且有效降低信噪比,对提高温室管理的智能化水平具有借鉴价值。
Aiming at the problems of high dimensionality and strong redundancy of greenhouse environmental data,which lead to low compression ratio and high peak signal-to-noise ratio in data processing,a high-dimensional data compression method for greenhouse environmental factors based on self coding neural network high-order feature extraction is proposed.By applying the improved regression equations,the missing values in the greenhouse environmental factor data are filled.Addressing the phenomenon of internal covariate transfer in deep self coding neural networks,an adaptive balance layer is added,combined with the small batch gradient descent method,a deep adaptive balanced self coding neural network is constructed to extract the high order features of greenhouse environmental factors,and based on vector quantization ideas,the relative errors are determined.By implementing new codebook calculations,the centroids of each partition are obtained.Based on the codebook training results,a highdimensional data compression method is designed.The results show that when the data volume exceeds 50 GB,the compression ratio of the designed method decreases by 0.7 percentage points,representing a 3.8%reduction.The overall compression performance is excellent.The peak signal-to-noise ratio does not significantly decrease with the increase in sampling rate,only dropping by 4 dB,representing a 7.5%reduction.The compression of the peak signalto-noise ratio has better reconstruction fidelity.In summary,the method has higher compression ratio and effective reduction of signal-to-noise ratio,which is valuable for improving the intelligent level of greenhouse management.
作者
冷令
王琳
吕金洪
李浩欣
吴伟斌
高婷
Leng Ling;Wang Lin;LüJinhong;Li Haoxin;Wu Weibin;Gao Ting(Zhongshan Polytechnic,Zhongshan,528400,China;Zhongshan Technical Secondary School,Zhongshan,528458,China;South China Agricultural University,Guangzhou,510642,China)
出处
《中国农机化学报》
北大核心
2026年第1期252-257,共6页
Journal of Chinese Agricultural Mechanization
基金
广东省普通高校重点领域专项科技服务乡村振兴项目(2023ZDZX4133)
中山市社会公益与基础研究专项项目(2024B2026)
广东省(深圳)数智农服产业园建设(FNXM012022020—1—03)。
关键词
改进回归方程
自编码神经网络
高阶特征提取
温室环境因子
高维数据压缩
improving regression equations
self coding neural network
high order feature extraction
greenhouse environmental factors
high dimensional data compression
