روشي براي اعتبارسنجي اطلاعات معابر در OSM بدون استفاده از اطلاعات مرجع و با استفاده از ساير اطلاعات OSM

A Validation Approach for OSM Roads Information without Using Authorized Information Based on the Other OSM Information

كنترل و تضمين كيفيت اطلاعات مكاني داوطلبانه يك از بزرگترين چالش هاي پيش روي اين نوع از گردآوري اطلاعات مكاني است. روش­ هاي ارائه شده براي ارزيابي كيفيت اطلاعات مكاني داوطلبانه به صورت كلي شامل دو رويكرد مقايسه‌ اين اطلاعات با يك مرجع معتبر و يا روش­ هاي بدون نياز به مقايسه با مرجع معتبر مي­شود. روش­ هاي گروه اول نيازمند دسترسي به اطلاعات مرجع هستند كه دسترسي به اين اطلاعات همواره امكان­پذير نمي­باشد. روش­ هاي گروه دوم، يا مثل روش اعتبارسنجي اطلاعات سعي بر به دست آوردن اعتبار اطلاعات بر اساس معيارهاي بدست آمده از خود اطلاعات دارند و يا با استفاده از قوانين حاكم بر داده­ ها مثل سازگاري منطقي داده­ ها به ارزيابي كيفيت داده ­ها مي­پردازند. در اين تحقيق تلاش بر ارائه روشي براي ارزيابي كيفيت اطلاعات مكاني داوطلبانه با استفاده از خصوصيات قابل محاسبه از داده مكاني و بدون نياز به مقايسه با اطلاعات مرجع گرديده است. اين تحقيق با استفاده از خصوصيات قابل محاسبه از داده­ هاي موجود در شبكه راه­هاي OpenStreetMap، به صورت خاص به تحليل ميزان دقت دسته­­ بندي خيابان ­ها، به عنوان بخشي از دقت معنايي مربوط به شبكه راه­ ها، مي­پردازد. در اين تحقيق از درخت تصميم و شبكه عصبي با ساختار چند لايه براي يادگيري اين خصوصيات استفاده گرديد. براي تعيين ميزان دقت دسته­ بندي بدست آمده، معيار­هاي دقت و بازخواني براي هر روش محاسبه شد. درخت تصميم با دقت وزن­ دار بدست آمده 84٫1٪ نشان­ دهنده يك مدل مناسب براي اين روش است. اين روش بر اساس اطلاعات قابل استخراج از داده ­ها استوار است و مي­تواند براي داده­ هاي هر شهر ديگري تعميم و استفاده گردد.

The geospatial data quality validation is building an effort to develop quality standards for geospatial information. The growth of Volunteer Geospatial Information (VGI) raises many issues regarding quality control or quality assurance. The lack of metadata and standards combined with an unknown motivation and used credibility lead to heterogeneous datasets with unknown quality. Thus, two main approaches have been suggested for assessing the quality of VGI: first approach is based on comparing VGI with authorized data and second approach tries to validate VGI without using authorized data. The methods based on first approach need the authorized information and this information is not always accessible or in case of existence might be expensive. Moreover, the method with global efficiency is not reachable. The methods based on second approach generally are three types: reviewing and correction by volunteers, using the general rules of geospatial data and extracting information from VGI to validate itself. The methods for reviewing and correction by volunteers argue that crowdsourcing data converges on the truth if people have the opportunity to review and correct errors. But features that many people have an interest will be more accurate than features that are of interest only to a few. In second methods, general rules for geospatial data like logical consistency used for VGI validation. The last method tries to use information from VGI to validate itself. For example a method determines information trust just by using the data history like production, change or removing of data to validate VGI just with itself. In this article we try to present a method for validation of quality of VGI without using authorized data and by using of other VGI. In this article, calculable characteristics of OpenStreetMap (OSM) data, as one of the important sources of VGI, are used for validate the quality of its VGI data. Specially, we analysis the accuracy of classification of OSM roads network as part of semantic data of VGI. Roads, such as motorway and residential, have different design and characteristics according to their function in street network. It is tried to classify these roads by finding and learning these distinguish characteristics of each road class. Machine learning models with decision tree and neural network algorithms are used to learn roads characteristics from OSM street network data. Decision tree and artificial neural network with multilayer perceptron are usable for data that contains errors and we have to assume VGI data always have errors. These two Statistical analysis, precision and recall are used for assessing final models. In accordance with the result, Decision tree model have 84.1% weighted average accuracy and represent a suitable model for this method. These methods are based on extractable information from VGI and could be used for any street network to classify the streets.