Computer vision tasks, such as image classification, semantic segmentation, and super resolution, are broadly utilized in many applications. Recent studies revealed that machine learning-based models for the computer ...
详细信息
ISBN:
(数字)9781665490627
ISBN:
(纸本)9781665490627
Computer vision tasks, such as image classification, semantic segmentation, and super resolution, are broadly utilized in many applications. Recent studies revealed that machine learning-based models for the computer vision tasks are vulnerable to adversarial attacks. Since the adversarial attack can disturb the computer vision models in real-world systems, many countermeasures have been proposed against the adversarial attacks, such as denoising, resizing, and machine learning-based super resolution model as a preprocessing. Recently, a prior work demonstrated that the super resolution model as a preprocessing can be vulnerable to the adversarial attack targeted to the preprocessing itself, only when the perturbation is inactive before the preprocessing. However, we also found that the perturbation before the preprocessing can be another serious threat if the super resolution model is used for a mitigation of adversarial attacks. In this paper, we propose Layered Adversary Generation (LAG) that generates the adversarial example by recursively injecting noises to clean image in white-box environment. We then show that LAG is effective to attack a semantic segmentation model even if the super resolution models with/without two countermeasures as auxiliary methods such as resizing and denoising are adopted to mitigate the adversarial attacks. Furthermore, we demonstrate that LAG is transferable across other super resolution models. Lastly, we discuss our attack method in gray-box and black-box environments, and suggests a mitigation for robust preprocessing.
patternrecognition is an important way to improve the quality of life and promote scientific research. through the big data captured by the three-axis sensor, this paper focuses on the establishment of relevant model...
详细信息
ISBN:
(纸本)9798350313444
patternrecognition is an important way to improve the quality of life and promote scientific research. through the big data captured by the three-axis sensor, this paper focuses on the establishment of relevant models for the accurate classification of 19 motion modes , and analyses of the generalization ability and overfitting. this paper establishes DeepConvLSTM model by combining the original deep learning convolutional neural network (CNN) and long- term short-term memory recursion (LSTM). the model code about DeepConvLSTM was written by Python and relevant data was imported. In terms of methods, this paper evaluates the generalization ability of the model by using setaside method and K-fold cross-validation method respectively from the algorithm level. In terms of indicators, this paper selects accuracy, accuracy rate, recall rate and F1 value. Finally, the conclusion is drawn as follows: when epoch increases to 40, Train loss and Val loss basically fall together, indicating that the accuracy and stability of the model are *** paper use logistic regression model to solve the classification problem of human behavior. It is found that the classification effect of DeepConvLSTM model was better than that of LR model with higher accuracy and no overfitting phenomenon. in addition, Recall is used to analyze the sensitivity of the model. After several tests, it is found that the model shows good sensitivity and stability when the training times reaches 60. On this basis, considering the cost and efficiency of training comprehensively, the number of training is set as 60 in this paper, and the sensitivity and robustness of the model are both good. In conclusion, this paper establishes a scientific action patternrecognition model according to the conditions of the subject and the data given. the model is simple and easy to popularize. After verification and analysis, the model in this paper has strong accuracy, robustness and sensitivity, and has certain pra
High-utility pattern mining, the process of searching highly relevant patterns in databases. pattern mining has gained popularity as a study area due to its many appliance in fields like bioinformatics, text mining, p...
详细信息
In recent years, the channel state information (CSI) feedback model based on artificial intelligence has emerged as a significant research area for 6G pre-research. However, existing methods mainly rely on the data-dr...
详细信息
In recent years, due to the importance of underwater image enhancement in underwater robot, underwater vehicle and ocean engineering, more and more extensive research has been done. It has evolved from implementing ph...
详细信息
the proceedings contain 34 papers. the topics discussed include: CASQ: enhancing human-object interaction detection via supplementary semantic information for interaction queries;TwinLiteNet: an efficient and lightwei...
ISBN:
(纸本)9798350327410
the proceedings contain 34 papers. the topics discussed include: CASQ: enhancing human-object interaction detection via supplementary semantic information for interaction queries;TwinLiteNet: an efficient and lightweight model for drivable area and lane segmentation in self-driving cars;efficient finetuning large language models for Vietnamese chatbot;comparative study of object detection models for abnormality detection on spinal x-ray images;CE-OST: contour emphasis for one-stage transformer-based camouflage instance segmentation;memory-driven region contrast for enhanced polyp semantic segmentation;intelligent attendance system: combining fusion setting with robust similarity measure for face recognition;attention pooling for beta wavelet filters in anomaly classification with graph neural network;and AANet: motorcycle ReID using multi-atrous convolution and self-attention mechanisms.
In this paper, we focus on the follicular unit registration problem for hair transplantation surgery robots based on binocular stereo vision system. the follicular units in both images are detected with YOLO V5 networ...
详细信息
In this paper, we propose an attack perception capability assessment method based on the specific semantic search for the problem of attack perception capability assessment of security protection equipment for power m...
详细信息
We all have such learning experiences that our teachers told us that there were no two identical leaves in the world. Based on this, this paper proposes that since there are no two identical leaves in the world, so as...
详细信息
Scene text recognition has historically concentrated on English, with limited advancements in developing solutions that perform well across multiple languages. Previous efforts in multilingual scene text recognition h...
详细信息
暂无评论