Research And Implementation Of Malware Detection Technology Based On Deep Learning

With the rapid development of Internet and Informatization of daily life,people rely on the internet more and more sticky.Along with the advent of 5G era and ever-changing in artificial intelligence and other technologies,cyber threats are still increasing,among which malware is one of the main concerns.The rapid evolution of malware has made traditional cybersecurity products insufficient to achieve promising detection accuracy.In order to alleviate malware attacks,research on malware detection is important.There are many drawbacks of current malware detection techniques.Firstly,existing malware detection techniques have uneven detection accuracy,in feature engineering,the influence of irrelevant features cannot be ignored,and it is easy to ignore factors such as the correlation between features and the dependency between sample attributes,which leads to insufficient depth of feature extraction.In addition,the performance of the model varies,some rely too much on hardware result in the system performance requirements are demanding,some detection accuracy cannot reach the standard,some models are too complicated to achieve ease-ofuse purpose.To address the problem of low detection accuracy,we propose a malware detection model based on deep learning in this paper.A detection method of LightGBM+1D-CNN is proposed firstly,whose innovation lies in applying LightGBM for feature extraction and selection,and sorted selecting features as input of 1D-CNN based on the importance.The LightGBM+1D-CNN achieve accuracy of 76.66%.In order to undermine the influence of irrelevant features on the detection results as much as possible,and to explore the correlation between features and the dependency between them,we introduce the self-attention mechanism and contrast learning in the LightGBM+1D-CNN approach(referred as LSCC).LSCC pays attention to the malicious samples after LightGBM feature extraction through self-attention mechanism as well as correlation within the features.LSCC focus more on the relationship and similarity between features,identification of attention points,identification of positive and negative samples.1D-CNN apply contrast loss function,and uniformity and alignment metrics to improve the classification accuracy effectively while prevent the occurrence of overfitting.The application of selfattention mechanism and contrast learning improves detection accuracy to 80.01%,which is 3.35%higher than LighGBM+1D-CNN.At the same time,we design and implement a deep learning-based malware detection system in this paper.The system operates with users through a visual interface.LSCC proposed in this paper is applied as the malware detection model in the system.Starting with uploading suspicious samples,the system will determine whether the detected samples are malware or not,and save detection results for access by other users afterwhile.