Deep learning hyperspectral image classification model-based sparse adversarial attack method

• Main Authors: YIN ZHAOXIA, SU HANG, TANG LICHUN, KONG CONG, LUO BIN
• Format: Patent
• Language: Chinese; English
• Published: 17.11.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a sparse adversarial attack method for a hyperspectral image classification model based on deep learning. The method comprises the following steps: selecting a specific waveband subset to generate a specific spectral waveband image based on a waveband selection algorithm of an attention mechanism; a generator of the sparse wave band disturbance generation module generates adversarial disturbance according to the specific spectral wave band image, a discriminator judges whether an input sample is a clean sample or an adversarial sample, and the disturbance generation level of the generator is improved through adversarial of the generator and the discriminator; generating adversarial wave band data by using a sparse wave band disturbance generation module; and combining the clean wave band data with the adversarial wave band data, recovering the clean wave band data and the adversarial wave band data into an original data size to obtain a sparse wave band adversarial sample, inputting th