Effective forward and backward propagation method for pixel-wise convolutional neural network

Satellite Image is used to analysis the ground area to identify the object. The main barrier of this image processing method is the pixel-wise analysis. The fully convolutional neural network is a popular method to process this type of image processing. However, train stage is taken the maximum computation of this method. Parallelize version of this method to reduce computation time is researching topics, now. But little work is done on optimize the complexity of the algorithm of this method. In this paper, we find an optimized algorithm solution of this algorithm. Also, the result of the parallel version of the proposed algorithm is discussed.