由於目前科技發展快速，使用深度學習影像自動辨識鋪面破壞快速崛起，起初以自動辨識各項鋪面破壞為主，現況將破壞態樣計算分數並開始導入鋪面狀況指標(Pavement Condition Index, PCI)，以達到道路巡查過程能夠紀錄並判斷每筆道路破壞狀況。但由於各項鋪面破壞態樣辨識率並無查證標準，故研究相關辨識驗證方法，並依相關標準訂定檢測之設備標準。依照相關辨識率做巡檢車之查證、校正及驗證，將檢測數據規格統一。 透過巡查車及檢測車各項優點並合併執行及運用，如巡查車能夠透過每日在各級單位之道路做巡視檢查，並將此巡查車輛安裝需檢測之設備，並根據檢測鋪面各項指標之設備做分級，如道路平整度、鋪面狀況指標、鋪面抗滑度、鋪面結構能力評估等相關檢測數據。依照相關國際設備分級標準做分析，並將人口、設備、鋪面破壞類型做分級。後續將回傳最新道路檢測數據，以達到各項指標能夠有最新檢測數據，並利用大數據分析方式，預測鋪面未來成效評估之狀況。再將檢測之數據自動回傳至鋪面檢測系統（Pavement Detection System， PDS），由相關檢測系統判斷各級路段之檢測分數，後續再將相關巡查數據及所有檢測數據於檢測平台中檢視及操作，最後回饋至鋪面管理系統。當檢測數據能夠更即時並有效存於平台後，也能以大數據方式分析每路段需維修之比較及排序，再依據每年度有的預算去編列能夠維修之路段，以達到鋪面生命週期之利用。 後續建議將駕駛巡檢車之人員做統一教育訓練，並將相關業主及業者辦理鋪面精進研討會，提升相關從業人員對於鋪面之專業認知，以及提升道路維修之品質。;Since the rapid development of technology, using the Deep Learning to automatically identify pavement damage for image is rapidly emerging. At the beginning, the automatic identification was only used by pavement damages. However, the application of pavement condition index (PCI) was also introduced later.In order to achieve road inspections and record the inspection process. Each road data. I hope that through this research, we can explore the combined use of various advantages of inspection vehicles and inspection vehicles. Such as IRI, PCI, SN and other related index data. When the data is more quantified, the dispatching system can also be incorporated into the paving inspection system to effectively predict the status of the paving inspection. sent back to the pavement detection system. The relevant system will determine the detection scores of all levels of road sections, and then the relevant inspection data and all the detection data will be reviewed in the detection platform. This research will also discuss related data issues. For example, these PCI identification programs or systems do not have certain specifications and standards. Therefore, I hope that this research can explore how to establish a unified standard and apply for relevant test data based on this standard. For certification, it is more objective to confirm all identification-related data with a third-party unit.
