Optimized CNN Learning Model With Multi‐Threading for Forgery Feature Detection in Real‐Time Streaming Approaches

Summary

The concept of internet of things connects everyday consequences to the internet in real time which improves our quality of life and experiences. Processing such real‐time data make memory and computations intensive. To acquire efficient outcome, real‐time deep‐learning convolutional neural network models can overcome challenging tasks related to video analyzing on edge devices. Attaining the robustness of the detection of content manipulation is made possible through multi‐threading, which is a sequence entropy with neural network model, to extract the features. The purpose of a convolutional neural network is to analyze and extract the specified required in‐depth features. The PRELU activation function is used to make a conversion of the input signal layer into an output signal in the neural network model. This produces the desired outcome with better convergence properties. To resolve small model performance levels caused by under‐fitting with high bias, a regularization technique can be applied to the model. Further increasing the work speed of the detection algorithm on edge devices can be done using the multi‐threading technique. Modular median filter is one of the widely used digital filter methods worldwide for the removal of noise generated to images or signals. Filtering is applied to the image processing in order to preserve the edges even after removing the noise. Finally, the model detects content forgery quickly and embeds it on the IoT end devices in order to produce fast real‐time inference on the edge device networks.

The overall model aims to minimize the computational resources, which further minimize the energy consumption of the edge devices with execution time, which is directly proportional to the speed of the network.