摘要
为提升我国水果质量检测水平,提出一种基于全子带栈式稀疏自编码的水果图像融合方法。首先利用滑动窗口技术将红外与可见光水果图像分块。接着,利用图像小块训练栈式稀疏自编码网络。然后,利用平移不变剪切波将待融合图像分解为低频子带和高频子带。低频子带与高频子带先利用滑动窗口技术提取子带分块,接着对每一对子带分块利用训练好的网络进行编码,并提出基于稀疏自编码的融合规则来实现子带分块的融合。最后,利用滑动窗口逆变换生成融合后水果图像。实验证明该方法在辅助水果质量检测的应用中是可行的、有效的。
Proposes a Stacked Sparse AutoEncoders （SSAE） of all subbands based fruit image fusion method to detect the quality of fruits. First, source images are divided into image blocks by sliding windows technology to train a SSAE network. Second, source images are decom- posed into low frequency subbands and high frequency subbands by Shift Invariant Shearlet Transform. Next, all frequency subbands are divided into subband blocks by sliding windows technology to learn codes by the SSAE network trained in the first step. Then, proposes two SSAE based fusion rules to fuse the frequency subbands. At last, the composite subbands are converted into fused image by the in- verse shift invariant shearlet transform. Experimental results show the proposed method is sufficient and efficient in the application of fruit quality inspection.
