TWI625174B - Method for detecting fracture of steel strip tail - Google Patents
Method for detecting fracture of steel strip tail Download PDFInfo
- Publication number
- TWI625174B TWI625174B TW106125333A TW106125333A TWI625174B TW I625174 B TWI625174 B TW I625174B TW 106125333 A TW106125333 A TW 106125333A TW 106125333 A TW106125333 A TW 106125333A TW I625174 B TWI625174 B TW I625174B
- Authority
- TW
- Taiwan
- Prior art keywords
- steel strip
- image
- tail end
- statistical
- descriptors
- Prior art date
Links
Landscapes
- Image Analysis (AREA)
Abstract
一種鋼帶尾端破損的偵測方法包含:擷取鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之多個統計特性描述子;從統計特性描述子當中篩選出多個具鑑別力的統計特性描述子;從鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之具鑑別力之統計特性描述子當中建立預測模型;對行進中的鋼帶進行尾端影像判斷步驟以獲得鋼帶尾端未知狀態影像;擷取鋼帶尾端未知狀態影像所相應之具鑑別力之統計特性描述子;以及利用預測模型來預測鋼帶是否有發生尾端破損。 A method for detecting damage at the end of a steel strip comprises: extracting a plurality of statistical characteristic descriptors corresponding to each of a normal state image of the tail end of the steel strip and a damaged state image of the tail end of the steel strip; and filtering from the statistical characteristic descriptor A plurality of discriminative statistical characteristic descriptors are generated; a prediction model is established from a statistical characteristic description of the discriminative force corresponding to each of the normal state image of the tail end of the steel strip and the image of the broken end state of the steel strip; The steel strip is subjected to the tail end image judging step to obtain an image of the unknown state at the end of the steel strip; the statistical characteristic descriptor of the discrimination force corresponding to the image of the unknown state at the end of the steel strip is taken; and the prediction model is used to predict whether the steel strip is used There is a broken end.
Description
本揭露是有關於一種鋼帶尾端破損的偵測方法,且特別是有關於一種以影像為基礎之鋼帶尾端破損的偵測方法。 The disclosure relates to a method for detecting the damage of the tail end of a steel strip, and in particular to a method for detecting the damage of the tail end of an image-based steel strip.
在目前的鋼帶生產過程中,鋼胚經加熱爐加熱後,會經過粗軋延和精軋延步驟,以獲得具有所需厚度之鋼帶。然而,在生產線的軋延製程中,鋼帶可能會產生偏移(side walk),進而造成鋼帶尾段破損等異常缺陷。在鋼帶進入軋延機具時,鋼帶尾端破損會對軋延機具的軋輥造成衝擊,而使軋輥表面形成凹凸不平的缺陷。此凹凸不平的缺陷會使得軋輥在軋延下一鋼帶時,將軋輥表面的缺陷轉印至鋼帶上,如此將導致後續經此軋輥軋延的鋼帶帶面形成缺陷而被剔退。這種因鋼帶尾端軋延異常而造成軋輥表面缺陷轉印的異常稱為輥軋轉印(Tail Pinch)。為了解決上述問題,需要一種方法來偵測鋼帶尾端是否發生破損。 In the current steel strip production process, after the steel embryo is heated by the heating furnace, the rough rolling and finishing rolling steps are carried out to obtain a steel strip having a desired thickness. However, in the rolling process of the production line, the steel strip may have a side walk, which may cause abnormal defects such as breakage of the steel strip tail. When the steel strip enters the rolling implement, the damage of the tail end of the steel strip will impact the roll of the rolling implement, and the surface of the roll will be uneven. This uneven defect causes the roll to transfer the defect on the surface of the roll to the steel strip when the next steel strip is rolled, which will cause the strip of the strip which is subsequently rolled by the roll to form a defect and be rejected. This abnormality in the transfer of the surface defects of the rolls due to the abnormal rolling at the end of the steel strip is called Tail Pinch. In order to solve the above problem, a method is needed to detect whether the end of the steel strip is broken.
本揭露的目的是在於提供一種以影像為基礎之鋼帶尾端破損的偵測方法,先對已知的鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像進行分析以建立預測模型,再對於行進中的鋼帶進行取像,當取得鋼帶尾端影像後,透過先前建立之預測模型來預測鋼帶尾端是否發生破損。 The purpose of the disclosure is to provide an image-based detection method for the damage of the tail end of a steel strip, and firstly analyze the normal state image of the tail end of the steel strip and the image of the broken end state of the steel strip to establish a prediction model. Then, the steel strip in progress is imaged, and after the image of the tail end of the steel strip is obtained, the previously established prediction model is used to predict whether the tail end of the steel strip is broken.
根據本揭露之上述目的,提出一種鋼帶尾端破損的偵測方法,包含:獲得多張鋼帶尾端正常狀態影像與多張鋼帶尾端破損狀態影像;進行影像處理步驟以擷取鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之多個統計特性描述子(descriptor);進行篩選步驟以從統計特性描述子當中篩選出多個具鑑別力的統計特性描述子;進行建模步驟以從鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之具鑑別力之統計特性描述子當中建立預測模型;對行進中的鋼帶進行尾端影像判斷步驟以獲得鋼帶尾端未知狀態影像;進行影像處理步驟以擷取鋼帶尾端未知狀態影像所相應之具鑑別力之統計特性描述子;以及進行預測步驟以利用預測模型來預測鋼帶是否有發生尾端破損。 According to the above object of the present disclosure, a method for detecting the damage of the tail end of a steel strip is provided, which comprises: obtaining a normal state image of a plurality of steel strip tail ends and an image of a broken state of a plurality of steel strip tail ends; performing an image processing step to extract steel A plurality of statistical property descriptors corresponding to each of the image of the normal state of the tail end and the image of the broken end state of the steel strip; performing a screening step to screen a plurality of discriminative statistics from the statistical property descriptors Characterization descriptor; performing a modeling step to establish a prediction model from the statistical characteristic description of the discriminative force corresponding to each of the normal state image of the steel strip tail end and the broken end state image of the steel strip; Carrying a tail image judgment step to obtain an image of an unknown state at the end of the steel strip; performing an image processing step to extract a statistical characteristic descriptor corresponding to the discriminative force of the image of the unknown end of the steel strip; and performing a prediction step to utilize the prediction The model is used to predict whether the steel strip is damaged at the end.
在一些實施例中,上述鋼帶尾端正常狀態影像、鋼帶尾端破損狀態影像與鋼帶尾端未知狀態影像之每一者為灰階影像。 In some embodiments, each of the normal state image of the tail end of the steel strip, the image of the broken end state of the steel strip, and the image of the unknown state of the tail end of the steel strip are gray scale images.
在一些實施例中,上述影像處理步驟包含:利用預設灰階閥值來將灰階影像轉換成二值化影像以區分出前景與背景;利用前景來求出多個邊界輪廓點;以及對邊界 輪廓點進行分析運算以求出統計特性描述子。 In some embodiments, the image processing step includes: converting a grayscale image into a binarized image to distinguish a foreground and a background by using a preset grayscale threshold; using the foreground to determine a plurality of boundary contour points; boundary The contour points are subjected to an analysis operation to find a statistical property descriptor.
在一些實施例中,上述影像處理步驟更包含:將灰階影像分割成多個區塊;以及對每一區塊進行灰階統計以求出每一區塊之區塊平均灰階與區塊灰階標準差,其中區塊平均灰階與區塊灰階標準差為統計特性描述子之其中二者。 In some embodiments, the image processing step further includes: dividing the grayscale image into a plurality of blocks; and performing grayscale statistics on each of the blocks to determine a block average grayscale and block of each block. Gray standard deviation, where the block average gray level and the block gray level standard deviation are two of the statistical characteristic descriptors.
在一些實施例中,上述篩選步驟包含:對統計特性描述子進行變異數分析;以及利用預設信心水準來篩選出具鑑別力的統計特性描述子。 In some embodiments, the screening step includes: performing a variance analysis on the statistical property descriptor; and screening the statistical property descriptor with the discriminative power using a preset confidence level.
在一些實施例中,上述建模步驟包含:對具鑑別力的統計特性描述子進行正規化處理;對已進行正規化處理之具鑑別力的統計特性描述子進行二進制編碼;以及進行線性迴歸運算來計算出關係函式。 In some embodiments, the modeling step includes: normalizing the statistical characteristic descriptor with discriminative power; performing binary encoding on the discriminative statistical characteristic descriptor that has been normalized; and performing linear regression operation To calculate the relationship function.
在一些實施例中,上述尾端影像判斷步驟包含以下步驟:(S5a)對行進中的鋼帶進行取像以獲得鋼帶影像;(S5b)利用鋼帶影像來定義出感興趣區域(region of interest,ROI);(S5c)對感興趣區域進行平均灰階值變化比較,以判斷鋼帶影像是否相應於鋼帶的尾端;若判斷結果為非,重複步驟(S5a)至(S5c);若判斷結果為是,進行步驟(S5d);以及(S5d)取鋼帶影像為鋼帶尾端未知狀態影像。 In some embodiments, the tail image determining step includes the following steps: (S5a) taking an image of the traveling steel strip to obtain a steel strip image; (S5b) using the steel strip image to define a region of interest (region of Interest, ROI); (S5c) comparing the average gray scale value change of the region of interest to determine whether the steel strip image corresponds to the tail end of the steel strip; if the judgment result is negative, repeat steps (S5a) to (S5c); If the result of the determination is YES, the step (S5d) is performed; and (S5d) the image of the steel strip is taken as an image of an unknown state at the end of the steel strip.
在一些實施例中,上述鋼帶沿著第一方向行進,且感興趣區域於第一方向上具有相對之第一側與第二側,其中上述平均灰階值變化比較包含以下步驟:計算第一側於垂直於第一方向之第二方向上的灰階值平均為第一平 均灰階值;計算第二側於第二方向上的灰階值平均為第二平均灰階值;以及當第一平均灰階值與第二平均灰階值的差值大於預設灰階差異閥值,則代表步驟(S5c)的判斷結果為是。 In some embodiments, the steel strip travels along a first direction, and the region of interest has a first side and a second side opposite the first direction, wherein the comparing the average grayscale value changes comprises the following steps: The gray level value of one side in the second direction perpendicular to the first direction is the first flat a gray scale value; calculating a gray scale value of the second side in the second direction to be a second average gray scale value; and when the difference between the first average gray scale value and the second average gray scale value is greater than a preset gray scale The difference threshold indicates that the judgment result of the step (S5c) is YES.
在一些實施例中,上述預測步驟包含:對鋼帶尾端未知狀態影像所相應之具鑑別力之統計特性描述子進行正規化處理;利用關係函式來對已進行正規化處理之具鑑別力的統計特性描述子進行運算以求得預測值;以及若預測值大於預設門檻值,則代表鋼帶有發生尾端破損。 In some embodiments, the predicting step comprises: normalizing the statistical characteristic descriptor of the discriminative force corresponding to the image of the unknown state at the end of the steel strip; and using the relational function to discriminate the normalized processing The statistical property descriptor performs an operation to obtain a predicted value; and if the predicted value is greater than a preset threshold value, it represents that the steel strip has a tail end breakage.
在一些實施例中,當鋼帶有發生尾端破損時,更發出警訊。 In some embodiments, a warning is issued when the steel belt is damaged at the end.
為讓本揭露的上述特徵和優點能更明顯易懂,下文特舉實施例,並配合所附圖式作詳細說明如下。 The above described features and advantages of the present invention will be more apparent from the following description.
1000‧‧‧偵測方法 1000‧‧‧Detection method
S1-S7、S2a-S2c、S5a-S5d‧‧‧步驟 S1-S7, S2a-S2c, S5a-S5d‧‧‧ steps
101、102‧‧‧感興趣區域 101, 102‧‧‧ Areas of interest
101A、102A‧‧‧第一側 101A, 102A‧‧‧ first side
101B、102B‧‧‧第二側 101B, 102B‧‧‧ second side
AF、AC、aC‧‧‧面積 A F , A C , a C ‧ ‧ area
PF、PC‧‧‧總長 P F , P C ‧‧‧ total length
(XF,YF)、(uF,vF)、(UF,VF)、(xF,yF)‧‧‧座標點 (X F , Y F ), (u F , v F ), (U F , V F ), (x F , y F ) ‧ ‧ punctuation
θp‧‧‧前景影像正彎角 θ p ‧‧‧ foreground image positive angle
θn‧‧‧前景影像負彎角 θ n ‧‧ ‧ foreground image negative angle
Li,1、Li,2‧‧‧前景影像彎角的邊長 L i,1 ,L i,2 ‧‧‧The length of the corner of the foreground image
從以下結合所附圖式所做的詳細描述,可對本揭露之態樣有更佳的了解。需注意的是,根據業界的標準實務,各特徵並未依比例繪示。事實上,為了使討論更為清楚,各特徵的尺寸都可任意地增加或減少。 A better understanding of the aspects of the present disclosure can be obtained from the following detailed description taken in conjunction with the drawings. It should be noted that, according to industry standard practices, the features are not drawn to scale. In fact, in order to make the discussion clearer, the dimensions of each feature can be arbitrarily increased or decreased.
[圖1]係繪示根據本揭露實施例之鋼帶尾端破損的偵測方法的流程圖。 FIG. 1 is a flow chart showing a method for detecting a broken end of a steel strip according to an embodiment of the present disclosure.
[圖2]係繪示根據本揭露實施例之影像處理步驟的流程圖。 FIG. 2 is a flow chart showing steps of image processing according to an embodiment of the present disclosure.
[圖3a]-[圖3c]係例示圖式用以說明根據本揭露實施例之多個用以描述影像外型特性之統計特性描述子。 [Fig. 3a] - [Fig. 3c] are diagrams for illustrating a plurality of statistical characteristic descriptors for describing image appearance characteristics according to an embodiment of the present disclosure.
[圖3d]係例示圖式用以說明根據本揭露實施例之多個用以描述影像灰階特性之統計特性描述子。 [Fig. 3d] is a schematic diagram for explaining a plurality of statistical characteristic descriptors for describing grayscale characteristics of an image according to an embodiment of the present disclosure.
[圖4]係繪示根據本揭露實施例之尾端影像判斷步驟的流程圖。 FIG. 4 is a flow chart showing a tail end image determining step according to an embodiment of the present disclosure.
[圖5]係繪示一示範例以說明感興趣區域之平均灰階值變化比較。 [Fig. 5] An exemplary embodiment is shown to illustrate the comparison of the average grayscale value variation of the region of interest.
以下仔細討論本發明的實施例。然而,可以理解的是,實施例提供許多可應用的概念,其可實施於各式各樣的特定內容中。所討論、揭示之實施例僅供說明,並非用以限定本發明之範圍。關於本文中所使用之『第一』、『第二』、...等,並非特別指次序或順位的意思,其僅為了區別以相同技術用語描述的元件或操作。 Embodiments of the invention are discussed in detail below. However, it will be appreciated that the embodiments provide many applicable concepts that can be implemented in a wide variety of specific content. The examples discussed and disclosed are illustrative only and are not intended to limit the scope of the invention. The terms "first", "second", "etc." used in this document are not intended to mean the order or the order, and are merely to distinguish between elements or operations described in the same technical terms.
在本揭露中,於鋼帶的多個軋機站之間設置影像擷取裝置,影像擷取裝置從鋼帶上方擷取鋼帶影像,影像擷取裝置電性連接至電腦裝置,以利用電腦裝置進行影像處理來分析每一軋機站之間的鋼帶影像。 In the present disclosure, an image capturing device is disposed between a plurality of rolling mill stations of the steel strip, the image capturing device picks up the steel strip image from above the steel strip, and the image capturing device is electrically connected to the computer device to utilize the computer device Image processing was performed to analyze the image of the steel strip between each rolling station.
圖1係繪示根據本揭露實施例之鋼帶尾端破損的偵測方法1000的流程圖。於步驟S1,獲得多張鋼帶尾端正常狀態影像與多張鋼帶尾端破損狀態影像。應注意的是,於步驟S1所獲得之鋼帶影像乃是灰階影像且是鋼帶尾端狀 態已知(鋼帶尾端正常或鋼帶尾端破損)之鋼帶尾端影像,因此於後續步驟即可針對鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之間的特性差異來進行分析。 FIG. 1 is a flow chart showing a method 1000 for detecting a broken end of a steel strip according to an embodiment of the present disclosure. In step S1, a normal state image of the tail end of the plurality of steel strips and a broken state image of the end of the plurality of steel strips are obtained. It should be noted that the image of the steel strip obtained in step S1 is a gray scale image and is a tail end of the steel strip. The state of the steel strip is known (the tail end of the steel strip is normal or the tail end of the steel strip is broken), so the difference between the normal state image of the tail end of the steel strip and the damage state of the tail end of the steel strip can be obtained in the subsequent steps. To analyze.
於步驟S2,進行影像處理步驟以擷取鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之多個統計特性描述子(descriptor)。圖2係繪示根據本揭露實施例之影像處理步驟S2的流程圖。首先,於步驟S2a,利用預設灰階閥值來將從步驟S1所獲得的灰階影像轉換成二值化影像以區分出前景與背景。在本揭露中,影像轉換的方式乃是先對灰階影像進行直方圖統計,並接著計算預設灰階閥值,其中,預設灰階閥值的計算方法可以是利用歐蘇法(Otsu Method)、最大熵(Maximal entropy)方法或者K均值(k-means)方法。然後,將灰階影像中影像灰階值超過預設灰階閥值的像素設為前景,反之則設為背景,從而將灰階影像轉換成二值化影像。 In step S2, an image processing step is performed to extract a plurality of statistical property descriptors corresponding to each of the normal state image of the steel strip tail end and the steel strip tail end damage state image. FIG. 2 is a flow chart showing an image processing step S2 according to an embodiment of the present disclosure. First, in step S2a, the grayscale image obtained in step S1 is converted into a binarized image by using a preset grayscale threshold to distinguish the foreground from the background. In the present disclosure, the image conversion method is to first perform histogram statistics on the grayscale image, and then calculate a preset grayscale threshold value, wherein the calculation method of the preset grayscale threshold value may be using the Ou Su method (Otsu Method), Maximum entropy method or K-means method. Then, the pixels in the grayscale image whose image grayscale value exceeds the preset grayscale threshold are set as the foreground, and vice versa, as the background, thereby converting the grayscale image into the binarized image.
於步驟S2b,利用前景來求出多個邊界輪廓點,其中,邊界輪廓點可用以描繪出前景影像輪廓。接著,於步驟S2c,對邊界輪廓點進行分析運算以求出多個統計特性描述子,其中,統計特性描述子可用以描述影像外型特性或用以描述影像灰階特性。表1列出本揭露之多個用以描述影像外型特性之統計特性描述子。圖3a至圖3c係例示圖式用以說明根據本揭露實施例之多個用以描述影像外型特性之統計特性描述子。 In step S2b, a plurality of boundary contour points are obtained using the foreground, wherein the boundary contour points can be used to depict the foreground image contour. Next, in step S2c, the boundary contour points are subjected to an analysis operation to obtain a plurality of statistical property descriptors, wherein the statistical property descriptors may be used to describe image appearance characteristics or to describe image grayscale characteristics. Table 1 lists a number of statistical characterization descriptors for describing image appearance characteristics of the present disclosure. 3a to 3c are diagrams for explaining a plurality of statistical characteristic descriptors for describing an image appearance characteristic according to an embodiment of the present disclosure.
請參照圖3a,其中灰底部分即為前景影像,在本發明的實施例中,以AF來表示前景影像面積,以(XF,YF)來表示前景影像中心座標,以PF來表示前景影像邊界長度(邊界總長),以(uF,vF)來表示前景影像邊界距前景影像中心最近點座標,以(UF,VF)來表示前景影像邊界距前景影像中心最遠點座標。因此,前景影像邊界不規則度(irregularity)相當於 Referring to Figure 3a, wherein the duplex portion is the foreground image, in the embodiment of the present invention, in order to represent the foreground image A F area to (X F, Y F) to represent the foreground image coordinates of the center, to be P F Indicates the foreground image boundary length (total length of the boundary), and (u F , v F ) represents the nearest point coordinate of the foreground image boundary from the foreground image center, and (U F , V F ) indicates that the foreground image boundary is farthest from the foreground image center. Point coordinates. Therefore, the foreground image boundary irregularity is equivalent to
請參照圖3b,其中虛線部分即為凸包影像,凸包影像的定義為可包含前景影像所有點集合的最小面積的凸多邊形。在本發明的實施例中,以AC來表示凸包影像面積,以(xF,yF)來表示凸包影像中心座標,以PC來表示凸包影像邊界長度(邊界總長)。因此,前景影像與凸包影像面積差之比例相當於,前景影像與凸包影像邊界長度差之比例相當於,前景影像緊密度(compactness)相當於,前景影像與凸包影像差影像區域面積aC相當於AC-AF,前景影像與凸包影像差影像區域中心Y座標相當於yF-YF。 Referring to FIG. 3b, the dotted line portion is a convex hull image, and the convex hull image is defined as a convex polygon that can include the smallest area of all the point sets of the foreground image. In the embodiment of the present invention, the convex envelope image area is represented by A C , the convex envelope image center coordinates are represented by (x F , y F ), and the convex envelope image boundary length (total length of the boundary) is represented by P C . Therefore, the ratio of the area difference between the foreground image and the convex hull image is equivalent to The ratio of the difference between the foreground image and the convex envelope image boundary length is equivalent to , foreground image compactness (compactness) is equivalent The foreground image and the convex hull image difference image area area a C is equivalent to A C -A F , and the foreground image and the convex hull image difference image area center Y coordinate are equivalent to y F -Y F .
請參照圖3c,前景影像具有多個前景影像,彎角其中以θp表示前景影像正彎角,以θn表示前景影像負彎角,以Li,1、Li,2表示前景影像彎角與前景影像輪廓線重疊的像素數目。應注意的是,在本實施例中,前景影像正彎角θp與前景影像負彎角θn為正向角,即逆時針方向為為角度增加的正方向。前景影像正彎角θp的定義為角度大於45度且Li,1與Li,2大於3像素的前景影像彎角,前景影像負彎角θn的定義為角度小於-45度且Li,1與Li,2大於3像素的前景影像彎角。值得一提的是,為了簡化標示,圖3c僅示出一個前景影像正彎角θp與一個前景影像負彎角θn,實際上,前景影像彎角數量可能更多。在本發明的實施例中,以Np來表示前景影像邊界正彎角個數,以Nn來表示前景影像邊界負彎角個數,以θmax來表示前景影像邊界最大正彎角角度,以θmin來表示前景影像邊界最小負彎角角度,以來表示前景影像邊界平均正彎角角度,以來表示前景影像邊界平均負彎角角度。因此,前景影像邊界最大正彎角彎曲能量相當於,前景影像邊界最小負彎角彎曲能量相當於,前景影像邊界平均正彎角彎曲能量相當於,前景影像邊界平均負彎角彎曲能量相當於
表2列出本揭露之多個用以描述影像灰階特性之統計特性描述子。圖3d係例示圖式用以說明根據本揭露實施例之多個用以描述影像灰階特性之統計特性描述子。 Table 2 lists a number of statistical feature descriptors used to describe the grayscale characteristics of the image. FIG. 3d is a schematic diagram for explaining a plurality of statistical characteristic descriptors for describing grayscale characteristics of an image according to an embodiment of the present disclosure.
請參照圖3d,其中灰底部分即為灰階影像。原始影像頂部平均灰階μT相當於灰階影像輪廓線內位於頂部1/5高度範圍內的原始像素平均灰階,原始影像底部平均灰階μB相當於灰階影像輪廓線內位於底部1/5高度範圍內的原始像素平均灰階,原始影像中央平均灰階μC相當於灰階影像輪廓線內排除頂部1/5與底部1/5高度範圍以外的原始像素平均灰階。因此,原始影像頂部/底部與中央平均灰階差值相當於max(μB-μC,μT-μC)。 Please refer to FIG. 3d, in which the gray bottom portion is a gray scale image. The average gray level μ T of the original image is equivalent to the original gray level of the original pixel within the height of the top 1/5 of the gray image contour. The average gray level μ B of the original image is equivalent to the gray image contour line at the bottom 1 The original pixel average grayscale in the /5 height range, the original image center average grayscale μ C is equivalent to the grayscale image outline excluding the top grayscale of the original pixel outside the top 1/5 and bottom 1/5 height range. Therefore, the difference between the top/bottom of the original image and the center average gray scale is equivalent to max (μ B -μ C , μ T -μ C ).
在本實施例中,影像處理步驟S2更包含:將灰階影像分割成多個區塊;以及對每一區塊進行灰階統計以求出每一區塊之區塊平均灰階與區塊灰階標準差,其中區塊平均灰階與區塊灰階標準差為統計特性描述子之其中二者。請 再次參照圖3d與表2,在本實施例中,將灰階影像分割為3×3個區塊,但本發明不限於此。接著,對每一區塊進行灰階統計以求出每一區塊之原始影像區塊平均灰階μj與原始影像區塊灰階標準差σj。 In this embodiment, the image processing step S2 further includes: dividing the grayscale image into a plurality of blocks; and performing grayscale statistics on each of the blocks to determine the average grayscale and the block of each block. Gray standard deviation, where the block average gray level and the block gray level standard deviation are two of the statistical characteristic descriptors. Referring to FIG. 3d and Table 2 again, in the present embodiment, the grayscale image is divided into 3×3 blocks, but the present invention is not limited thereto. Then, gray scale statistics are performed on each block to find the original image block average gray level μ j of each block and the original image block gray level standard deviation σ j .
請回到圖1,於步驟S3,進行篩選步驟以從統計特性描述子當中篩選出多個具鑑別力的統計特性描述子。在本揭露中,採用變異數分析(analysis of variance,ANOVA)方法來計算各個統計特性描述子是否有顯著性。在本揭露中,對每一統計特性描述子而言分別有兩類(群)資料,即鋼帶尾端正常與鋼帶尾端破損,分別計算每一統計特性描述子之兩類(群)資料之平均組內變異(within-group variability)與平均組間變異(between-group variability),則變異數分析方法當中的F值(F value)=(平均組間變異)/(平均組內變異)。接著定義出信心水準(confidence level),例如95%,若某一統計特性描述子的F值超過上述定義之信心水準,則表示該統計特性描述子為具有顯著性,亦即,該統計特性描述子為具有鑑別力的統計特性描述子。 Returning to FIG. 1, in step S3, a screening step is performed to filter a plurality of discriminative statistical property descriptors from the statistical property descriptors. In the present disclosure, an analysis of variance (ANOVA) method is used to calculate whether each statistical property descriptor is significant. In the present disclosure, there are two types of (group) data for each statistical property descriptor, that is, the normal end of the steel strip and the tail end of the steel strip are broken, and two types of statistical characteristic descriptors are respectively calculated (group). The mean-group variability and the average-group variability of the data, the F-value ((mean-group variation)/(mean-group variation) in the variance analysis method ). Then define a confidence level, for example 95%. If the F value of a statistical property descriptor exceeds the confidence level defined above, it indicates that the statistical property descriptor is significant, that is, the statistical property description The child is a statistical feature descriptor with discriminative power.
請再次參照圖1,於步驟S4,進行建模步驟以從鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像之每一者所相應之具鑑別力之統計特性描述子當中建立預測模型。在上述的建模步驟中,先對各個具鑑別力的統計特性描述子進行正規化處理,其中,正規化處理的目的是將具鑑別力的統計特性描述子之尺度(scale)範圍調整到一個合理範 圍,例如尺度範圍為0~1或-1~+1,從而利於後續運算並確保後續建模品質。在本揭露中,正規化處理之尺度範圍調整的方式可為等比例縮放或是標準差標準化(zero-mean normalization)(即減去平均數再除以標準差)。在上述的建模步驟中,接著對已進行正規化處理之具鑑別力的統計特性描述子進行二進制編碼並進行線性迴歸運算來計算出關係函式。 Referring again to FIG. 1, in step S4, a modeling step is performed to establish a prediction model from a statistical description of the discriminative statistical characteristics corresponding to each of the normal state image of the steel strip tail end and the broken end state image of the steel strip end. . In the above modeling step, each of the discriminative statistical characteristic descriptors is first normalized, wherein the purpose of the normalization processing is to adjust the scale of the discriminative statistical characteristic descriptor to one Reasonable For example, the scale ranges from 0 to 1 or -1 to +1, which facilitates subsequent operations and ensures subsequent modeling quality. In the present disclosure, the scale adjustment of the normalization process may be a scaling or zero-mean normalization (ie, subtracting the average and dividing by the standard deviation). In the above modeling step, the discriminative statistical property descriptors that have been normalized are then binary coded and linear regression operations are performed to calculate the relationship function.
以下藉由一示範例來說明如何利用線性迴歸運算以計算出關係函式,其中,具有鑑別力的統計特性描述子的數量為n個,於步驟S1當中,鋼帶尾端正常狀態影像的數量為i張,鋼帶尾端破損狀態影像的數量為j張,且i+j=m。第1張鋼帶尾端正常狀態影像的第1個具有鑑別力的統計特性描述子的數值為a1,1,第i張鋼帶尾端正常狀態影像的第n個具有鑑別力的統計特性描述子的數值為ai,n,第1張鋼帶尾端破損狀態影像的第1個具有鑑別力的統計特性描述子的數值為ai+1,1,第j張鋼帶尾端破損狀態影像的第n個具有鑑別力的統計特性描述子的數值為am,n。因此可利用矩陣Am×n來表示具有鑑別力的統計特性描述子,並利用矩陣Bm×1來進行二進制編碼,其中0代表鋼帶尾端正常,1代表鋼帶尾端破損,則矩陣Bm×1當中的0共有i個,1共有j個,矩陣Am×n如下式(1)所示,矩陣Bm×1如下式(2)所示:
透過線性迴歸(linear regression)運算可知矩陣Am×n與矩陣Bm×1的關係式為Am×n*Xn×1=Bm×1,故計算關係函式Xn×1的運算式為Xn×1=(ATA)-1(ATB)。 Through the linear regression operation, the relationship between the matrix A m × n and the matrix B m × 1 is A m × n * X n × 1 = B m × 1 , so the operation of the relational function X n × 1 is calculated. The formula is X n × 1 = (A T A) -1 (A T B).
應注意的是,在本揭露之步驟S4當中的建模步驟並不限定使用線性迴歸方法來進行建模,建模的方法也可以是利用類神經網路(artificial network)或支持向量機器(support vector machine)等迴歸方法。 It should be noted that the modeling step in step S4 of the present disclosure does not limit the use of linear regression methods for modeling, and the modeling method may also utilize an artificial neural network or a support vector machine (support). Vector machine) and other regression methods.
請再次參照圖1,於步驟S5,對行進中的鋼帶進行尾端影像判斷步驟以獲得鋼帶尾端未知狀態影像。圖4係繪示根據本揭露實施例之尾端影像判斷步驟S5的流程圖。於步驟S5a,對行進中的鋼帶進行取像以獲得鋼帶影像。於步驟S5b,利用鋼帶影像來定義出感興趣區域(region of interest,ROI),由於在鋼帶影像中,鋼帶的寬度可能不會維持一致,感興趣區域的高度必須讓鋼帶的寬度之最窄處可以含括,而感興趣區域的長度相當於影像擷取裝置可以在鋼帶行進方向上拍攝到的總長。於步驟S5c,對感興趣區域進行平均灰階值變化比較,以判斷鋼帶影像是否相應於鋼帶的尾端。 Referring to FIG. 1 again, in step S5, the trailing end image judging step is performed on the traveling steel strip to obtain an image of an unknown state at the end of the steel strip. FIG. 4 is a flow chart showing a tail end image determining step S5 according to an embodiment of the present disclosure. In step S5a, the traveling steel strip is imaged to obtain a steel strip image. In step S5b, the steel strip image is used to define a region of interest (ROI). Since the width of the steel strip may not be consistent in the steel strip image, the height of the region of interest must be the width of the steel strip. The narrowest portion can be included, and the length of the region of interest is equivalent to the total length that the image capturing device can take in the direction of travel of the steel strip. In step S5c, an average grayscale value change comparison is performed on the region of interest to determine whether the steel strip image corresponds to the trailing end of the steel strip.
以下藉由一示範例來說明步驟S5c是如何進行的,圖5係繪示一示範例以說明感興趣區域之平均灰階值變 化比較。應注意的是,由於經軋延之鋼帶呈現紅熱狀態,故於鋼帶影像中代表鋼帶者,即圖5當中灰色區塊者,具有較高的灰階值,反之,鋼帶影像中並非代表鋼帶者,實質上為暗色(黑色),故具有較低的灰階值。如圖5所示,感興趣區域101、102於鋼帶行進方向上具有相對之第一側101A、102A與第二側101B、102B,計算感興趣區域101、102之第一側101A、102A於垂直於鋼帶行進方向上之每一像素點的灰階值平均為第一平均灰階值,計算感興趣區域101、102之第二側101B、102B於垂直於鋼帶行進方向上之每一像素點的灰階值平均為第二平均灰階值。當鋼帶影像為鋼帶尾端影像,如圖5左圖所示,代表感興趣區域101之第一側101A與第二側101B會有明顯的灰階值落差,因此可定義預設灰階差異閥值,例如100灰階,當第一平均灰階值與第二平均灰階值的差值大於預設灰階差異閥值,即可知鋼帶影像為鋼帶尾端影像。當鋼帶影像並非鋼帶尾端影像,如圖5右圖所示,代表感興趣區域102之第一側102A與第二側102B不會有明顯的灰階值落差,亦即,當第一平均灰階值與第二平均灰階值的差值沒有大於預設灰階差異閥值,即可知鋼帶影像並非鋼帶尾端影像。值得一提的是,如圖5左圖所示,當感興趣區域101之第一側101A與第二側101B有明顯的灰階值落差,且感興趣區域101之第一側101A的灰階值低於感興趣區域101之第二側101B,則可知鋼帶影像為鋼帶尾端。反之,若感興趣區域之第一側與第二側有明顯的灰階值落差,且感興趣區域之第一側的灰階值高於感興趣區域之第 二側,則鋼帶影像為鋼帶頭端。而在本揭露的步驟S5c中,應判斷鋼帶影像是否相應於鋼帶尾端。 The following describes an example of how step S5c is performed. FIG. 5 illustrates an exemplary example to illustrate the average grayscale value change of the region of interest. Comparison. It should be noted that since the rolled steel strip exhibits a red hot state, the steel strip representing the steel strip, that is, the gray block in Fig. 5, has a higher gray scale value, and vice versa, in the steel strip image. It is not representative of the steel strip, it is essentially dark (black), so it has a lower gray scale value. As shown in FIG. 5, the regions of interest 101, 102 have opposite first sides 101A, 102A and second sides 101B, 102B in the direction of travel of the steel strip, and the first sides 101A, 102A of the regions of interest 101, 102 are calculated. The gray scale value of each pixel point perpendicular to the traveling direction of the steel strip is averaged as a first average gray scale value, and the second sides 101B, 102B of the regions of interest 101, 102 are calculated to be perpendicular to the traveling direction of the steel strip. The grayscale value of the pixel is on average the second average grayscale value. When the steel strip image is the tail end image of the steel strip, as shown in the left figure of FIG. 5, the first side 101A and the second side 101B representing the region of interest 101 have a significant gray scale value difference, so the preset gray scale can be defined. The difference threshold, for example, 100 gray scale, when the difference between the first average gray scale value and the second average gray scale value is greater than the preset gray scale difference threshold, the steel strip image is the end image of the steel strip. When the steel strip image is not the end image of the steel strip, as shown in the right figure of FIG. 5, the first side 102A and the second side 102B representing the region of interest 102 do not have a significant gray scale value drop, that is, when the first The difference between the average grayscale value and the second average grayscale value is not greater than the preset grayscale difference threshold, and the steel strip image is not the end image of the steel strip. It is worth mentioning that, as shown in the left figure of FIG. 5, when the first side 101A and the second side 101B of the region of interest 101 have a significant gray scale value difference, and the gray scale of the first side 101A of the region of interest 101 The value is lower than the second side 101B of the region of interest 101, and the steel strip image is the end of the steel strip. Conversely, if the first side and the second side of the region of interest have significant grayscale value differences, and the grayscale value of the first side of the region of interest is higher than the region of interest On the two sides, the image of the steel strip is the end of the steel strip. In the step S5c of the present disclosure, it is determined whether the steel strip image corresponds to the tail end of the steel strip.
請回到圖4,若於步驟S5c判斷得知鋼帶影像並非鋼帶尾端影像,則重複步驟S5a至步驟S5c,亦即,繼續對行進中的鋼帶進行取像,直到步驟S5c的判斷結果為是。當步驟S5c的判斷結果為是,則進行步驟S5d,取鋼帶影像為鋼帶尾端未知狀態影像。 Referring back to FIG. 4, if it is determined in step S5c that the steel strip image is not the tail end image of the steel strip, step S5a to step S5c are repeated, that is, the taking of the steel strip in progress is continued until the judgment of step S5c is performed. The result is yes. When the result of the determination in the step S5c is YES, the process proceeds to a step S5d, and the image of the steel strip is taken as an image of the unknown state of the end of the steel strip.
請回到圖1,於步驟S6,進行影像處理步驟以擷取鋼帶尾端未知狀態影像所相應之具鑑別力之統計特性描述子。步驟S6的做法與步驟S2實質上相同,差別僅在於步驟S6是針對鋼帶尾端未知狀態影像且僅擷取具鑑別力之統計特性描述子,故在此不再贅述。 Returning to FIG. 1, in step S6, an image processing step is performed to extract a statistical characteristic descriptor of the discrimination force corresponding to the image of the unknown state at the end of the steel strip. The step S6 is substantially the same as the step S2. The only difference is that the step S6 is for the unknown state image of the tail end of the steel strip and only the statistical characteristic descriptor having the discriminating power is extracted, so it will not be described here.
請參照圖1,於步驟S7,進行預測步驟以利用預測模型來預測鋼帶是否有發生尾端破損。在上述的預測步驟中,先對鋼帶尾端未知狀態影像所相應之具鑑別力之統計特性描述子進行正規化處理,關於正規化處理已於上述步驟S4的敘述當中說明過,在此不再贅述。在上述的預測步驟中,接著利用關係函式來對已進行正規化處理之具鑑別力的統計特性描述子進行運算以求得預測值。 Referring to FIG. 1, in step S7, a prediction step is performed to predict whether the steel strip is damaged at the end end by using a prediction model. In the above-mentioned prediction step, the statistical characteristic descriptor of the discrimination force corresponding to the unknown state image of the tail end of the steel strip is first normalized, and the normalization processing has been described in the description of the above step S4, and Let me repeat. In the above-described prediction step, the relational function of the discriminative statistical characteristic that has been subjected to the normalization processing is then operated by the relational function to obtain the predicted value.
以下藉由一示範例來說明如何利用關係函式來求得預測值,其中,具有鑑別力的統計特性描述子的數量同樣為n個,鋼帶尾端未知狀態影像的第1個具有鑑別力的統計特性描述子的數值為c1,鋼帶尾端未知狀態影像的第n個具有鑑別力的統計特性描述子的數值為cn,因此可利用矩陣 C1×n來表示鋼帶尾端未知狀態影像所相應之具有鑑別力的統計特性描述子,即矩陣C1×n=(c1…cn)。接著利用關係式b=C1×n*Xn×1來求得預測值b。 The following is an example to illustrate how to use the relational function to obtain the predicted value. The number of statistical characteristic descriptors with discriminative power is also n, and the first one of the unknown state image of the steel strip has the discriminative power. The value of the statistical characteristic descriptor is c 1 , and the value of the nth discriminative statistical characteristic descriptor of the unknown state image of the steel strip end is c n , so the matrix C 1×n can be used to represent the end of the steel strip. The statistical characteristic descriptor of the discriminative force corresponding to the unknown state image, that is, the matrix C 1 × n = (c 1 ... c n ). Next, the predicted value b is obtained by using the relation b = C 1 × n * X n × 1 .
在上述的預測步驟中,最後,定義預設門檻值,例如預設門檻值為0.5,若預測值大於預設門檻值,則代表鋼帶有發生尾端破損。除此之外,當鋼帶有發生尾端破損時,可透過例如驅動警報裝置來發出警訊,以通知現場人員鋼帶有發生尾端破損,從而使現場人員可採取適當措施來避免因為鋼帶尾端破損而造成軋輥表面缺陷轉印的異常。 In the above prediction step, finally, the preset threshold value is defined, for example, the preset threshold value is 0.5, and if the predicted value is greater than the preset threshold value, the steel strip is damaged at the end. In addition, when the steel belt is damaged at the end, the warning device can be used to send a warning to notify the site personnel that the steel belt is damaged at the end, so that the field personnel can take appropriate measures to avoid the steel. Abnormality in the transfer of the surface defects of the roll due to breakage at the end.
由上述可知,本揭露之鋼帶尾端破損的偵測方法,先對已知的鋼帶尾端正常狀態影像與鋼帶尾端破損狀態影像進行分析以建立預測模型,再對於行進中的鋼帶進行取像,當取得鋼帶尾端影像後,透過先前建立之預測模型來預測鋼帶尾端是否發生破損。 It can be seen from the above that the method for detecting the damage of the tail end of the steel strip of the present invention first analyzes the normal state image of the tail end of the steel strip and the image of the broken end state of the steel strip to establish a prediction model, and then for the steel in progress. The belt is taken for imaging, and after the image of the tail end of the steel strip is obtained, the previously established prediction model is used to predict whether the tail end of the steel strip is broken.
以上概述了數個實施例的特徵,因此熟習此技藝者可以更了解本揭露的態樣。熟習此技藝者應了解到,其可輕易地把本揭露當作基礎來設計或修改其他的製程與結構,藉此實現和在此所介紹的這些實施例相同的目標及/或達到相同的優點。熟習此技藝者也應可明白,這些等效的建構並未脫離本揭露的精神與範圍,並且他們可以在不脫離本揭露精神與範圍的前提下做各種的改變、替換與變動。 The features of several embodiments are summarized above, and those skilled in the art will be able to understand the aspects of the disclosure. Those skilled in the art will appreciate that the present disclosure can be readily utilized as a basis for designing or modifying other processes and structures, thereby achieving the same objectives and/or achieving the same advantages as the embodiments described herein. . It should be understood by those skilled in the art that the invention may be made without departing from the spirit and scope of the disclosure.
Claims (8)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW106125333A TWI625174B (en) | 2017-07-27 | 2017-07-27 | Method for detecting fracture of steel strip tail |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW106125333A TWI625174B (en) | 2017-07-27 | 2017-07-27 | Method for detecting fracture of steel strip tail |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI625174B true TWI625174B (en) | 2018-06-01 |
TW201910021A TW201910021A (en) | 2019-03-16 |
Family
ID=63255669
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW106125333A TWI625174B (en) | 2017-07-27 | 2017-07-27 | Method for detecting fracture of steel strip tail |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI625174B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110245392A (en) * | 2019-05-28 | 2019-09-17 | 上海交通大学 | Clad failure trend prediction method based on artificial neural network |
TWI711801B (en) * | 2019-09-18 | 2020-12-01 | 中國鋼鐵股份有限公司 | Method for detecting the position of holes in strip product |
TWI749735B (en) * | 2019-09-12 | 2021-12-11 | 日商東芝三菱電機產業系統股份有限公司 | Reduction generation prediction system |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110280608A (en) * | 2019-07-19 | 2019-09-27 | 北京宇轩智能科技有限公司 | The intelligent accurate rolled piece tracking processing method of steel rolling |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101829689A (en) * | 2010-03-31 | 2010-09-15 | 北京科技大学 | Drift fault recognition method of hot-rolling strip steel based on sound signals |
CN102288613A (en) * | 2011-05-11 | 2011-12-21 | 北京科技大学 | Surface defect detecting method for fusing grey and depth information |
TW201500124A (en) * | 2013-06-24 | 2015-01-01 | China Steel Corp | Adaptive image adjustment method using base model |
CN105787486A (en) * | 2015-12-11 | 2016-07-20 | 昆明理工大学 | Method for detecting girder cracks based on image processing |
CN106952258A (en) * | 2017-03-23 | 2017-07-14 | 南京汇川图像视觉技术有限公司 | A kind of bottle mouth defect detection method based on gradient orientation histogram |
-
2017
- 2017-07-27 TW TW106125333A patent/TWI625174B/en active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101829689A (en) * | 2010-03-31 | 2010-09-15 | 北京科技大学 | Drift fault recognition method of hot-rolling strip steel based on sound signals |
CN102288613A (en) * | 2011-05-11 | 2011-12-21 | 北京科技大学 | Surface defect detecting method for fusing grey and depth information |
TW201500124A (en) * | 2013-06-24 | 2015-01-01 | China Steel Corp | Adaptive image adjustment method using base model |
CN105787486A (en) * | 2015-12-11 | 2016-07-20 | 昆明理工大学 | Method for detecting girder cracks based on image processing |
CN106952258A (en) * | 2017-03-23 | 2017-07-14 | 南京汇川图像视觉技术有限公司 | A kind of bottle mouth defect detection method based on gradient orientation histogram |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110245392A (en) * | 2019-05-28 | 2019-09-17 | 上海交通大学 | Clad failure trend prediction method based on artificial neural network |
CN110245392B (en) * | 2019-05-28 | 2022-12-09 | 上海交通大学 | Artificial neural network-based shell damage state prediction method |
TWI749735B (en) * | 2019-09-12 | 2021-12-11 | 日商東芝三菱電機產業系統股份有限公司 | Reduction generation prediction system |
TWI711801B (en) * | 2019-09-18 | 2020-12-01 | 中國鋼鐵股份有限公司 | Method for detecting the position of holes in strip product |
Also Published As
Publication number | Publication date |
---|---|
TW201910021A (en) | 2019-03-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
TWI625174B (en) | Method for detecting fracture of steel strip tail | |
CN115239704B (en) | Accurate detection and repair method for wood surface defects | |
Choi et al. | Detection of pinholes in steel slabs using Gabor filter combination and morphological features | |
CN105374015A (en) | Binary method for low-quality document image based on local contract and estimation of stroke width | |
CN113962997B (en) | Strip steel edge crack defect detection method and system based on image processing | |
CN104517276A (en) | Checker corner detection method | |
EP3493103A1 (en) | Human body gender automatic recognition method and apparatus | |
CN111667475B (en) | Machine vision-based Chinese date grading detection method | |
CN116563641A (en) | Surface defect identification method and system based on small target detection | |
CN112378350B (en) | Flatness detection method for PIN PIN of network transformer | |
CN115049657A (en) | Glass defect detection method | |
CN108020554A (en) | A kind of steel strip surface defect recognition detection method | |
CN116630304B (en) | Lithium battery mold processing detection method and system based on artificial intelligence | |
CN115953398A (en) | Defect identification method for strip steel surface | |
CN117078678B (en) | Waste silicon wafer shape detection method based on image recognition | |
CN114863492A (en) | Method and device for repairing low-quality fingerprint image | |
CN112508913A (en) | Cable section edge detection method based on image detection | |
CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
CN106530273B (en) | High-precision FPC (Flexible printed Circuit) linear line detection and defect positioning method | |
CN117314901B (en) | Scale-adaptive chip detection neural network system | |
Fang et al. | Towards real-time crack detection using a deep neural network with a Bayesian fusion algorithm | |
Huang et al. | Dynamic programming and connected component analysis for an enhanced pavement distress segmentation algorithm | |
Malakar et al. | Handwritten and printed word identification using gray-scale feature vector and decision tree classifier | |
CN115100510B (en) | Tire wear degree identification method | |
CN108898080B (en) | Ridge line neighborhood evaluation model-based crack connection method |