基于深度学习的化妆品塑料瓶缺陷检测

提出一种基于深度卷积神经网络的化妆品塑料瓶表面缺陷检测算法。采用百万像素级别的工业相机采集大量的塑料瓶图像样本,并通过HSV(hue,saturation,value)颜色空间变换和Otsu阈值分割等方法对图像进行预处理。采用随机图像变换法对数据集进行增强,并对图像进行标准归一化处理。在卷积神经网络模型中应用深度可分离卷积和Dropout技术以减少参数量,从而避免过度拟合。使用训练样本集训练该模型,并在测试集中将结果与几种经典图像识别算法进行比较分析,结果显示,本文算法的识别准确率高达约0.97。由此表明本文算法的效果优于其他经典算法,有望将其应用于化妆品塑料瓶缺陷检测的工业自动化系统,以提升缺陷识别效果,从而提高生产效率。

Defect Detection of Cosmetic Plastic Bottles Based on Deep Learning

This paper proposed a method for detecting surface defects of cosmetic plastic bottles based on deep convolutional neural networks.A megapixel industrial camera was used to collect a large number of plastic bottle image samples,and the images were preprocessed by HSV(hue,saturation,value)color space transformation and Otsu threshold segmentation.The data set was enhanced by random image transformation method and the image was normalized.Deep detachable convolution and Dropout techniques were applied in the convolutional neural network model to reduce the number of parameters and avoid over-fitting.The model was trained with the training sample set and the results were compared with several classical image recognition algorithms in the test set.The results show that the proposed algorithm achieves a recognition accuracy close to 0.97,indicating that the algorithm is better than other classical ones.Therefore,the proposed method can be applied to the industrial automation system for defect detection of cosmetic plastic bottles,so as to improve the recognition effect of defects and production efficiency.