| 基于语义分割的食品标签文本检测 |
| |
| 引用本文： | 田萱,王子亚,王建新.基于语义分割的食品标签文本检测[J].农业机械学报,2020,51(8):336-343. |
| |
| 作者姓名： | 田萱 王子亚 王建新 |
| |
| 作者单位： | 北京林业大学信息学院,北京100083;国家林业草原林业智能信息处理工程技术研究中心,北京100083;北京林业大学信息学院,北京100083 |
| |
| 基金项目： | 国家重点研发计划项目(2018YFC1603305、2018YFC1603302) |
| |
| 摘 要： | 食品包装上的标签文本含有生产日期、营养成分、生产厂家等食品相关信息,这些不仅为消费者购买食品提供了重要依据,也有助于食品监督抽检机构发现潜在的食品安全问题。食品标签文本检测是食品标签自动识别的前提,有助于降低人工录入成本、提高数据处理效率。基于食品包装图像构建数据集,提出了一种基于语义分割的距离场模型,以检测食品标签。该模型包含像素分类和距离场回归两类任务,其中像素分类任务分割处理图像中的文本区域,距离场回归任务预测文本区域内的像素点到该区域边界的归一化距离。为提升模型的检测性能,在回归预测模块中通过增加注意力模块优化模型结构,并针对距离场回归任务损失值过小、影响模型训练优化问题对其损失函数进行了改进。消融实验结果表明,增加注意力模块和损失函数的改进使得模型的准确率分别提高了4.39、3.80个百分点,有效提高了检测准确率。食品包装图像数据集的对比实验表明,采用本文模型检测食品标签文本具有较好的性能,其召回率、准确率分别达到87.61%、76.50%。
|
| 关 键 词： | 食品标签 语义分割 文本检测 文本识别 |
| 收稿时间： | 2020/5/13 0:00:00 |
Text Detection of Food Labels Based on Semantic Segmentation |
| |
| TIAN Xuan,WANG Ziy,WANG Jianxin.Text Detection of Food Labels Based on Semantic Segmentation[J].Transactions of the Chinese Society of Agricultural Machinery,2020,51(8):336-343. |
| |
| Authors: | TIAN Xuan WANG Ziy WANG Jianxin |
| |
| Institution: | Beijing Forestry University;Engineering Research Center for Forestry-oriented Intelligent Information Processing of National Forestry and Grassland Administration |
| |
| Abstract: | The label texts on food package include some information like production date, nutrition facts and production corporation etc. The information provides important foundation for consumers to buy food. It also can help the food supervision and inspection administrations to discover the potential problems of food safety. Food label detection is the groundwork of food label recognition. It can help to decrease the heavy workload of manual inputting and advance efficiency of data processing. The dataset of food label was constructed firstly, and then a semantic segmentation based distance field model (DFM) was proposed. In DFM two tasks were included: pixel classification and distance field regression. The pixel classification task was used to segment the text from background regions, and the distance field regression task was used to predict the normalized distance from the pixel located in the text region to the boundary of text region. For effectively using the correlation of two tasks, an attention module was added into DFM to optimize the model structure. In addition, the loss function was improved to resolve the loss value of the distance field regression as it was too small to train smoothly. The results of ablation experiment showed that the accuracy of the proposed model was increased by 4.39 percentage points and 3.80 percentage points respectively according to the improvement of attention module and loss function. The comparative experiments of different model methods showed that DFM had good performance in detecting the text of food labels, and the recall rate and precision were 87.61% and 76.50%, respectively. |
| |
| Keywords: | food labels semantic segmentation text detection text recognition |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
