基于卷积神经网络的色情图像检测

PORNOGRAPHIC IMAGE DETECTION BASED ON CONVOLUTIONAL NEURAL NETWORK

针对通常使用的色情图像检测方法中难以获取准确的色情图像特征的问题,提出一种以数据为导向基于深度卷积神经网络来获取图像特征的色情图像检测方法。对含色情内容和不含色情内容的图片数据集进行数据增强处理,接着使用Inception模块设计及建立卷积神经网络模型;使用批量随机梯度下降算法训练卷积神经网络获取色情图像特征;使用训练好的模型识别一张图像是否是色情图像。测试集检测正确率达到了99.06%,对比实验表明所设计的网络模型因其参数更少比其他模型更不易过拟合并比其他方法实现了更高的准确率。

Aiming at the problem that it is difficult to obtain the accurate pornographic image features in commonly used pornographic image detection methods, a data-oriented detection method of pornographic images based on depth convolution neural networks was proposed. Firstly, image data containing the pornographic content and the pornographic content were processed using data augmentation ; the number of neurons was determined according to the input and output and the network model with less parameters and higher ability using Inception module was designed. Then, a batch random gradient descent algorithm was used to train the convolution neural network to obtain the pornographic image features. Finally, it classified that whether an image was a pornographic image with the trained model. Test set detection accuracy reached 99.06%. Contrast tests showed that the designed network model is less prone to over-fitting and had higher accuracy than other methods because of its fewer parameters than other models.