




摘"要:精準(zhǔn)識(shí)別鐵路車號(hào)可以為煤廠裝車提供依據(jù),從而保證裝車環(huán)節(jié)高效順利地完成。為此,提出了基于深度信任網(wǎng)絡(luò)模型的烏東選煤廠鐵路車號(hào)圖像識(shí)別方法。首先,利用高速攝像機(jī)設(shè)備采集原始的車號(hào)圖像,并利用索貝爾算子檢測(cè)圖像邊界;然后,根據(jù)列車車號(hào)的字體筆畫寬度特點(diǎn),采筆畫寬度變換算法定位確定圖像中的車號(hào)區(qū)域,并利用LBP算法提取車號(hào)區(qū)域內(nèi)的特征;最后,將提取的特征輸入到深度信任網(wǎng)絡(luò)模型中,在訓(xùn)練網(wǎng)絡(luò)模型并不斷更新參數(shù)后,準(zhǔn)確識(shí)別車號(hào)圖像。實(shí)驗(yàn)表明:該方法能夠精準(zhǔn)識(shí)別烏東選煤廠鐵路列車車號(hào)圖像。在深度信任網(wǎng)絡(luò)模型中,當(dāng)受限玻爾茲曼機(jī)網(wǎng)絡(luò)為4層、隱含層節(jié)點(diǎn)個(gè)數(shù)為128個(gè)時(shí),該模型的分類識(shí)別能力最強(qiáng),訓(xùn)練損失最小,性能最佳。
關(guān)鍵詞:深度信任網(wǎng)絡(luò);邊界檢測(cè);車號(hào)定位;圖像識(shí)別;筆畫寬度變換;特征提取
中圖分類號(hào):TP751""""""文獻(xiàn)標(biāo)識(shí)碼:A
Image"Recognition"Method"of"Railway"Car"Number"of"Wudong"
Coal"Preparation"Plant"Based"on"Deep"Trust"Network"Model
WEI"Weijie1,"QI"Jian1,"ZHOU"Nan1,"LIU"Huanan1,"GAO"Huiying2
(1.National"Energy"Group"Xinjiang"Energy"Co.,"Ltd.,"washing"Center,Urumqi,"Xinjiang"830000,"China;
2.Tianjin"Meiteng"Technology"Co.,Ltd.,"Tianjin"300000,China)
Abstract:Accurate"identification"of"railway"vehicle"number"can"provide"basis"for"coal"plant"loading,"thus"ensuring"the"efficient"and"smooth"completion"of"the"loading"process."Therefore,"a"method"of"railway"vehicle"number"image"recognition"based"on"deep"trust"network"model"in"Wudong"Coal"Preparation"Plant"is"proposed."Firstly,"the"original"vehicle"number"image"is"collected"by"highspeed"camera"equipment,"and"the"image"boundary"is"detected"by"Sobel"operator;"Then,"based"on"the"font"stroke"width"characteristics"of"the"train"number,"a"stroke"width"transformation"algorithm"is"used"to"locate"and"determine"the"train"number"area"in"the"image,"and"the"LBP"algorithm"is"used"to"extract"features"within"the"train"number"area;"Finally,"the"extracted"features"are"input"into"the"deep"trust"network"model."After"training"the"network"model"and"constantly"updating"the"parameters,"the"vehicle"number"image"is"accurately"recognized."The"experiment"shows"that"this"method"can"accurately"recognize"the"train"number"image"of"Wudong"Coal"Preparation"Plant."In"the"deep"trust"network"model,"when"the"restricted"Boltzmann"network"is"4"layers"and"the"number"of"hidden"layer"nodes"is"128,"the"model"has"the"strongest"classification"recognition"ability,"the"minimum"training"loss"and"the"best"performance.
Key"words:deep"trust"network;"boundary"detection;"vehicle"number"positioning;"image"recognition;"stroke"width"change;"feature"extraction
烏東選煤廠坐落于新疆,是我國西部地區(qū)規(guī)模最大的動(dòng)力煤選煤廠[1,2]。在該煤廠原裝車系統(tǒng)中,已配備了RFID車號(hào)識(shí)別系統(tǒng),正常情況下可以滿足日常作業(yè)需求,但是車廂射頻識(shí)別IC卡消磁或RFID系統(tǒng)出現(xiàn)故障時(shí),則無法對(duì)煤廠鐵路車號(hào)進(jìn)行識(shí)別,進(jìn)而影響裝車作業(yè)進(jìn)度。另外在自動(dòng)化改造的大環(huán)境下,對(duì)鐵路車號(hào)實(shí)行自動(dòng)化識(shí)別改造也是一種趨勢(shì),所以提出另外一種鐵路車號(hào)圖像識(shí)別方法對(duì)原有RFID車號(hào)識(shí)別系統(tǒng)進(jìn)行補(bǔ)充非常必要[3-5]。
近年來,國內(nèi)外學(xué)者……