|Abstract: ||鑑於傳統上以人工方式進行鋪面破壞實地 調查,可能遭致危險及判斷不正確等缺失,本研 究嘗試利用影像處理及類神經網路識別方式, 建立自動鋪面破壞識別系統。經實證研究發現 正確率為81%,但將不當之光線因素造成之誤差 扣除,則正確率可達91%,而主要之誤差來源包括 散落鋪面之油漬、樹葉等,以及不同鋪設時期 造成之鋪面顏色相異等原因。另外研究過程中 亦發現國內外之鋪面破壞識別標準皆無一明確 之識別標準〔1〕〔2〕〔3〕,在進行鋪面維修 工作時,實有必要建立一標準之識別標準。|
Traditionally, the technology of pavement distress survey is manpower, and it may lead to such shortcomings as incorrect judgement, making workers dangerous, and so on. This paper presents work towards the use of employing neural network model of mask-based on image process for the automatic pavement crack recognition system. The demonstration have been proved that the rate of correctness is 81%. If we can avoid the error that due to inadequate shadow factor, the correctness can improve to 91%. The main sources of error include the leaves, the oil that spilt on pavement and different color based on different paved period,..., etc. And at the same time, we find it didn't have a clear-cut distress identification standard yet now, so it's surely necessary to own an identification standard in order to do pavement maintenance job.