Research on the Optimization Algorithm to Avoid Malicious Code Attacks

In order to solve the accuracy, time consuming and anti-detection ability of traditional LAN malicious attack code classification method, this paper proposes a LAN malicious code classification method based on improved optimization algorithm. This method adopts 3 D coordinate technology to expand the spatial location of malicious code to a wider range, so as to improve the characteristics of malicious code, so as to effectively prevent malicious attacks. To solve the time-consuming problem, a new transfer learning technology is adopted to improve the anti-confounding ability. In addition, in order to solve the problem of large network computing volume and slow code recovery speed, the MobileNetV3 convolutional neural network model is adopted, combined with the Ranger optimization algorithm technology, to speed up the time-consuming network convergence process. Both the CIFAR-10 and VisualQA datasets achieved 93.5% and 91.2%, with a 2.3% improvement in accuracy and a 25% improvement in detection efficiency compared to Sentiment140. The improved algorithm has significantly improved the accuracy in the classification than the traditional method, reaching the level of 1.05%. These improved algorithms not only improve the accuracy of the model, but also enhance the resistance to confounding, and improve the efficiency of classification detection.