Binary cross entropy with deep learning technique for Image classification

This paper discussed about a unique neural network approach stimulated by a technique that has reformed the field of computer vision: pixel-wise image classification, which we combine with binary cross entropy loss and pre training of the CNN (Convolutional Neural Network) as an auto encoder.The pixel-wise classification technique directly estimates the image source label for each timefrequency (T-F) bin in our image, thus eliminating common pre-and-post processing tasks.The proposed convolutional neural network is trained by using the binary mask as the target output label.The binary mask identifies the dominant image source in each T-F bin of the magnitude spectrogram of a mixture signal, by considering each T-F bin as a pixel with a multi-label (for each image source).Binary Cross Entropy is used as the training objective, so as to minimize the average probability error between the target and predicted label for each pixel.The Inception V3 architecture is used to further boost ImageNet classification accuracy.The results show that the proposed algorithm has the highest accuracy.