Method, system, and computer-readable storage medium for detecting a Trojan horse image

Abstract

The present invention relates to a Trojan horse picture detection method, system and computer-readable storage medium; the Trojan horse picture detection method includes: S1, decoding and segmenting pictures in an image data set; wherein, the image data set includes normal pictures and Trojan horse pictures; S2. Construct the overall features of the text based on the text samples obtained by segmentation; S3. Extract the features of the text vector through the two-way LSTM with attention and three parallel CNNs with different kernel_sizes, and obtain the LSTM features and CNN features; S4. Combine the three After that, through the fully connected layer and softmax activation function, a hybrid neural network model is established and weight training is performed; S5, the picture to be detected is decoded, segmented, and then processed to obtain three types of features and input weight training. The hybrid neural network model outputs the classification result of the picture to be detected. The invention improves the precision of Trojan horse detection.

Description

technical field [0001] The invention belongs to the technical field of Trojan horse detection, and in particular relates to a Trojan horse picture detection method, system and computer-readable storage medium. Background technique [0002] A large number of images are added to the Internet every year, but some images look normal, but actually hide Trojan horse codes. Using neural networks, Trojan horse images can be decoded, detected and classified to purify the network environment and ensure the healthy development of the Internet. [0003] In the prior art, the patent document with the publication number CN111723368A discloses a malicious code detection method based on Bi-LSTM and self-attention, which converts the malicious code into a sequence of byte streams of uniform length, and each byte element utilizes multi-dimensional One-hot encoding representation; the bidirectional long short-term memory Bi-LSTM model is used to automatically learn the sequence characteristics...

AI Technical Summary

Problems solved by technology

The text after the normal decoding of the picture is messy, and the codes of the Trojan horse program are various, so it is very difficult to identify. In the existing technology, feature identification is only performed on time series features, which easily leads to limited accuracy of image classification.

Images