A Novel Deep Learning Based Approach For Image Classification

To address the worsening problem, a deep residual learning technique is provided in this paper. Deeper neural networks are more difficult to train since there is less certainty that each stacked layer corresponds to the requisite underlying mapping. For training networks that are substantially deeper than previously employed networks, we proposed a residual learning technique. Instead of learning unreferenced functions, we deliberately reframe the layers to acquire residual functions with reference to the layer inputs. We offer thorough empirical evidence that indicates residual networks become more adaptable and precise as network depth rises. Also, investigate residual nets with depths of up to 152 layers on the ImageNet dataset, which is 8 levels deeper than VGG networks which has less complexity.