JP7207842B2

JP7207842B2 - Grain size determination method and system for ground material

Info

Publication number: JP7207842B2
Application number: JP2019082811A
Authority: JP
Inventors: 勝利藤崎; 佳克米丸; 健一川野; 出黒沼
Original assignee: Kajima Corp
Current assignee: Kajima Corp
Priority date: 2019-04-24
Filing date: 2019-04-24
Publication date: 2023-01-18
Anticipated expiration: 2039-04-24
Also published as: JP2020180835A

Description

本発明は地盤材料の粒度判定方法及びシステムに関し，とくに地盤材料の画像から粒度を判定する方法及びシステムに関する。 The present invention relates to a method and system for determining grain size of ground material, and more particularly to a method and system for determining grain size from an image of ground material.

ダム・堤防・路体・路盤・路床等の土木構造物を構築する際に，材料・施工の合理化を図る観点から，現場付近の地山等の採取場で調達された地盤材料（粘土，砂，礫等の様々な粒径の粒状材が混在する土木材料）を原材料として使用することがある。例えばＣＳＧ（ＣｅｍｅｎｔｅｄＳａｎｄａｎｄＧｒａｖｅｌ），セメント改良土等の構造材料は，現場付近で調達した地盤材料Ｓ（ＣＳＧ材，セメント改良土母材等と呼ばれる）に水及びセメントを混合してそのまま施工するものであり，大量且つ高速な施工を可能とする利点を有している。現場付近で調達した地盤材料Ｓを骨材としたコンクリートを構造材料として使用する場合もある。 Ground materials (clay, Civil engineering materials mixed with granular materials of various grain sizes such as sand and gravel) may be used as raw materials. For example, structural materials such as CSG (Cemented Sand and Gravel) and cement-improved soil are constructed by mixing water and cement with ground material S (called CSG material, cement-improved soil base material, etc.) procured near the site. It has the advantage of enabling large-scale and high-speed construction. In some cases, concrete using ground material S procured near the site as an aggregate is used as a structural material.

ＣＳＧ材，セメント改良土母材，骨材等の地盤材料Ｓは，図６に示すように，採取場１等で調達したのち破砕装置１ａ等で適当に破砕することはあるが，基本的に人為的な粒度調整を施さずそのまま用いるものであり，粒度や含水率の変動によって構造物の品質（とくに強度）に変動を生じさせる。そのため，採取場１からストックヤード２経由でトラック等の運搬機械３によって工事現場に順次搬入される地盤材料Ｓを適宜抜き取って粒度，含水率等を確認し，地盤材料Ｓと混合する水Ｗ及びセメントＣの添加量を調整して構造材料の品質を適切に管理することが求められる。図中の符号４は工事現場に搬入された地盤材料Ｓを投入するホッパーを示し，符号９は水Ｗ及びセメントＣを混合する混合装置（ミキサー等）を示す。 Ground materials S such as CSG materials, cement-improved soil base materials, aggregates, etc., as shown in Fig. 6, may be appropriately crushed by a crushing device 1a after being procured at a collection site 1, etc., but basically It is used as it is without artificially adjusting the particle size, and the quality (especially strength) of the structure will change due to the variation of the particle size and moisture content. Therefore, the ground material S, which is sequentially carried into the construction site by a transport machine 3 such as a truck from the sampling site 1 via the stockyard 2, is appropriately extracted, and the particle size, moisture content, etc. are checked, and the water W mixed with the ground material S and It is required to appropriately control the quality of the structural material by adjusting the amount of cement C added. Reference numeral 4 in the figure indicates a hopper for throwing in the ground material S carried into the construction site, and reference numeral 9 indicates a mixing device (mixer, etc.) for mixing water W and cement C. As shown in FIG.

一般に地盤材料Ｓの粒度（粒度分布）は，混在する各粒状材の粒径ｄを横軸（対数軸）とし，各粒径ｄ以下の粒状材の全体に対する質量百分率（＝その粒径ｄ以下の粒状材の総質量／粒状材全体の総質量×１００。以下，加積通過率ということがある）を縦軸（線形軸）とした片対数グラフ，すなわち図７のような粒径加積曲線Ｐによって表すことができる。粒径加積曲線Ｐを作成するためには，地盤材料Ｓを複数回篩い分けして複数の粒径ｄの加積通過率Ｐ（ｄ）を求める必要がある。しかし，地盤材料Ｓの篩い分けには非常に手間がかかるため，地盤材料Ｓの撒き出し画像Ｇからコンピュータ（画像処理装置）によって粒径加積曲線Ｐを求める方法が開発されている（特許文献１～３参照）。 In general, the particle size (particle size distribution) of the ground material S is expressed by the horizontal axis (logarithmic axis) of the particle size d of each mixed granular material, and the mass percentage (= the particle size d or less) of the total particle size d or less. The total mass of the granular material / total mass of the entire granular material × 100. Hereinafter, it may be referred to as the cumulative passage rate) is a semi-logarithmic graph with the vertical axis (linear axis), i.e. It can be represented by curve P. In order to create the grain size accumulation curve P, it is necessary to sift the ground material S a plurality of times and obtain the accumulated passage rate P(d) of a plurality of grain sizes d. However, since it takes a lot of time and effort to sieve the ground material S, a method has been developed to determine the particle size accumulation curve P from the unloaded image G of the ground material S using a computer (image processing device) (Patent document 1-3).

地盤材料Ｓは最大粒径から最小粒径までの範囲（粒径範囲）が広いので，地盤材料Ｓの撒き出し画像Ｇから特定の粒径ｄ（例えば１００ｍｍ）以下の粒状材を全て検出して加積通過率Ｐ（ｄ）を直接的に求めることは困難である。特許文献１～３は，画像Ｇから特定の粒径ｄ以上の粒状材を全て検出し，画像Ｇの全体面積Ｅに対する粒径ｄ以上の粒状材の面積ｅの総和（Σｅ）の割合Σｅ／Ｅ（以下，粒度インデクスＩという）から，その粒径ｄの加積通過率Ｐ（ｄ）を間接的に求めるものである。 Since the ground material S has a wide range from the maximum particle size to the minimum particle size (particle size range), all granular materials having a specific particle size d (for example, 100 mm) or less are detected from the spread image G of the ground material S. It is difficult to obtain the cumulative pass rate P(d) directly. Patent Documents 1 to 3 detect all granular materials having a specific particle size d or more from the image G, and calculate the ratio Σe/ From E (hereinafter referred to as particle size index I), the cumulative passage rate P(d) of the particle size d is obtained indirectly.

図８は，地盤材料Ｓの粒度インデクスＩと加積通過率Ｐ（ｄ）との変換式Ｋの一例を示すグラフである。このグラフは，地盤材料Ｓの複数の粒径ｄ＝１０ｍｍ，２０ｍｍ，３０ｍｍ，４０ｍｍについて従来の篩い分けで加積通過率Ｐ（ｄ）を求め，他方でその地盤材料Ｓの撒き出し画像Ｇから各粒径ｄ＝１０ｍｍ，２０ｍｍ，３０ｍｍ，４０ｍｍの粒度インデクスＩ（ｄ）を算出し，それらを二次平面にプロットして両者の関係を示したものである。同グラフから，各粒径ｄの加積通過率Ｐ（ｄ）＝ｙが粒度インデクスＩ（ｄ）＝ｘの多次元回帰モデルＫ（ｙ＝Σａ_ｎ・ｘ^ｎ）で表せること，すなわち各粒径ｄの粒度インデクスＩを加積通過率Ｐに変換できることが分かる。このような変換式Ｋ（例えば回帰モデル）を利用すれば，地盤材料Ｓの撒き出し画像Ｇから所要粒径ｄの粒度インデクスＩを求めて加積通過率Ｐに変換することできる。 FIG. 8 is a graph showing an example of the conversion formula K between the grain size index I of the ground material S and the cumulative passage rate P(d). This graph obtains the cumulative passage rate P(d) by conventional sieving for a plurality of particle sizes d = 10 mm, 20 mm, 30 mm, and 40 mm of the ground material S, and on the other hand, from the spread image G of the ground material S The particle size index I(d) for each particle size d=10 mm, 20 mm, 30 mm, and 40 mm was calculated and plotted on a quadratic plane to show the relationship between the two. From the same graph, it can be seen that the cumulative passage rate P(d)=y for each particle size d can be represented by the multidimensional regression model K (y=Σa _n x ⁿ ) of the particle size index I(d)=x. It can be seen that the particle size index I of the diameter d can be converted into an accumulated throughput P. By using such a conversion formula K (for example, a regression model), the particle size index I of the required particle size d can be obtained from the ground image G of the ground material S and converted into the cumulative passage rate P.

図９は，粒度インデクスＩを利用して地盤材料Ｓの粒径加積曲線Ｐ（ｄ）を作成する特許文献３の粒度計測システムを示す。図示例のシステムは，地盤材料Ｓの撒き出し画像Ｇを撮影するデジタルカメラ等の撮像装置５と，その画像Ｇを入力して地盤材料Ｓの粒径加積曲線Ｐ（ｄ）を作成するコンピュータ３０とを有している。図示例のコンピュータ３０は，キーボード等の入力装置３１と，ディスプレイ・プリンタ等の出力装置３２と，粒度インデクスＩを加積通過率Ｐに変換する変換式Ｋ等を記憶する記憶手段３５とを有する。 FIG. 9 shows the particle size measurement system of Patent Document 3, which uses the particle size index I to create the particle size accumulation curve P(d) of the ground material S. As shown in FIG. The system shown in the figure includes an imaging device 5 such as a digital camera that captures an image G of ground material S, and a computer that inputs the image G and creates a grain size accumulation curve P(d) of the ground material S. 30. The illustrated computer 30 has an input device 31 such as a keyboard, an output device 32 such as a display/printer, and a storage means 35 for storing a conversion formula K for converting the granularity index I into the cumulative pass rate P. .

また図示例のコンピュータ３０は，内蔵プログラム（画像処理プログラム）として，画像Ｇから地盤材料Ｓ中の各粒状材の輪郭を検出する検出手段４１と，各粒状材の輪郭から粒径ｄ及び面積ｅを求めて複数の粒径ｄの粒度インデクスＩ（ｄ）を算出する算出手段４２と，算出した複数の粒径ｄの粒度インデクスＩ（ｄ）を変換式Ｋにより加積通過率Ｐに変換して粒径加積曲線Ｐ（ｄ）を作成する作成手段４３と，作成した粒径加積曲線Ｐ（ｄ）に基づき地盤材料Ｓの粒度品質を判定・評価する評価手段４４とを有している。 Further, the computer 30 in the illustrated example includes, as a built-in program (image processing program), a detection means 41 for detecting the contour of each granular material in the ground material S from the image G, and a particle size d and an area e from the contour of each granular material. Calculation means 42 for calculating the particle size index I(d) of a plurality of particle sizes d by calculating the and an evaluation means 44 for judging and evaluating the grain size quality of the ground material S based on the created grain size accumulation curve P(d). there is

図９の粒度計測システムによれば，工事現場に搬入される地盤材料Ｓを適宜抜き取って撒き出し，その撒き出し画像Ｇを撮影してコンピュータ３０に入力することにより，地盤材料Ｓの粒径加積曲線Ｐ（ｄ）を迅速に作成することができ，地盤材料Ｓの粒度品質を例えば１５分～３０分に１回程度の頻度で判定・評価（管理）することができる。また，図５に示すように地盤材料Ｓを搬送するベルトコンベア１０上で振動させて撒き出す（分散させる）技術が開発されており（特許文献４参照），そのような振動ベルトコンベア１０等を用いて撒き出す時間を更に短縮ないし省略することにより，工事現場に搬入される地盤材料Ｓの粒度品質をほぼ連続的に判定・評価（管理）することも可能である。 According to the particle size measurement system of FIG. 9, the ground material S brought into the construction site is appropriately extracted and spread out, and the image G of the spread out is taken and input to the computer 30, whereby the grain size of the ground material S is increased. The product curve P(d) can be quickly created, and the grain size quality of the ground material S can be determined and evaluated (managed) at a frequency of once every 15 to 30 minutes, for example. In addition, as shown in FIG. 5, a technique for vibrating and sprinkling (dispersing) the ground material S on a belt conveyor 10 that conveys it has been developed (see Patent Document 4), and such a vibrating belt conveyor 10 etc. By further shortening or omitting the time for sprinkling using the method, it is possible to almost continuously determine and evaluate (manage) the grain size quality of the ground material S brought into the construction site.

図５（Ａ）は，地盤材料Ｓを載置して搬送しながら分散させる特許文献４の振動ベルトコンベア１０を示す。図示例のベルトコンベア１０は，通常のベルトコンベアと同様に駆動プーリ１１とテールプーリ１３との間に環状に帯状ベルト１４を架け渡したものである。駆動装置１２により駆動プーリ１１を駆動し，駆動プーリ１１とテールプーリ１３との間で帯状ベルト１４を回転させることにより，ベルト面に載置した地盤材料Ｓを搬送する。ベルト面の下方には，搬送方向に沿って複数のキャリアローラ１５，１６が並べられており，ベルト面の上流部のキャリアローラ１５は第１支持体２１によって連結支持され，下流部のキャリアローラ１６は第１支持体２１から縁切りされた第２支持体１７によって連結支持されている。第１支持体２１及び第２支持体１７は，それぞれ支持脚によって工事現場の基盤上に支持されている。 FIG. 5(A) shows a vibrating belt conveyor 10 of Patent Document 4 in which the ground material S is placed and distributed while being conveyed. The belt conveyor 10 shown in the figure has a belt-shaped belt 14 looped between a driving pulley 11 and a tail pulley 13, like a normal belt conveyor. By driving the driving pulley 11 by the driving device 12 and rotating the strip belt 14 between the driving pulley 11 and the tail pulley 13, the ground material S placed on the belt surface is conveyed. Below the belt surface, a plurality of carrier rollers 15 and 16 are arranged along the conveying direction. 16 is connected and supported by a second support 17 cut off from the first support 21 . The first support 21 and the second support 17 are each supported on the foundation of the construction site by supporting legs.

図５（Ａ）のベルトコンベア１０において，ベルト面の荷重はキャリアローラ１５，１６を介して第１支持体２１及び第２支持体１７により支持されるが，第１支持体２１と第２支持体１７とは相互に縁切りされているので，第１支持台２１にはキャリアローラ１５の荷重のみが伝達され，他のキャリアローラ１６の荷重は伝達されない。また，ベルトコンベア１０は第１支持体２１を振動させる振動装置２２を有しているが，振動装置２２の振動は第２支持体１７に伝達されることはなく，第１支持体２１とそれに連結されたキャリアローラ１５のみを振動させることができる。 In the belt conveyor 10 shown in FIG. 5(A), the load on the belt surface is supported by the first support 21 and the second support 17 via the carrier rollers 15 and 16. Since it is separated from the body 17, only the load of the carrier roller 15 is transmitted to the first support 21, and the load of the other carrier rollers 16 is not transmitted. In addition, the belt conveyor 10 has a vibrating device 22 for vibrating the first support 21, but the vibration of the vibrating device 22 is not transmitted to the second support 17. Only the coupled carrier roller 15 can be vibrated.

図５（Ａ）の撮像装置５は，ベルトコンベア１０の下流部の無振動エリアに，ベルト面と対向するように配置されている。ベルトコンベア１０の上流部において地盤材料Ｓを振動させながらベルト面画像Ｇを撮影すると，撮影のたびに地盤材料Ｓ中の粒子や土塊が異なる形として写り込み，画像Ｇから測定される粒度の精度が低下するおそれがある。図５（Ａ）のように，地盤材料Ｓを上流の振動エリアにおいて振動させて分散させたのち，その下流の無振動エリアにおいてベルト面画像Ｇを撮影することにより，画像Ｇから粒度を精度よく求めることができる。 The imaging device 5 in FIG. 5A is arranged in a non-vibration area downstream of the belt conveyor 10 so as to face the belt surface. When the belt surface image G is photographed while vibrating the ground material S in the upstream part of the belt conveyor 10, the particles and soil lumps in the ground material S appear as different shapes each time the image is taken, and the accuracy of the particle size measured from the image G. may decrease. As shown in FIG. 5(A), after the ground material S is vibrated and dispersed in the upstream vibrating area, the belt surface image G is photographed in the downstream non-vibrating area, so that the particle size can be accurately determined from the image G. can ask.

特開２０１０－２４９５５３号公報JP 2010-249553 A 特開２０１１－１６３８３６号公報JP 2011-163836 A 特開２０１３－２５７１８８号公報JP 2013-257188 A 特開２０１６－１２４６６５号公報JP 2016-124665 A

原田達也「機械学習プロフェッショナルシリーズ画像認識」講談社，２０１７年５月２４日発行Tatsuya Harada "Machine Learning Professional Series Image Recognition" Kodansha, May 24, 2017 涌井良幸・涌井貞美「Ｅｘｃｅｌでわかるディープラーニング超入門」技術評論社，２０１８年１月５日発行Yoshiyuki Wakui and Sadami Wakui, "Introduction to Deep Learning with Excel," Gijutsuhyoronsha, January 5, 2018

上述したように，例えば図５の振動ベルトコンベア１０等に地盤材料Ｓを載置して搬送しながら分散させ，ベルトコンベア１０上の分散した状態の地盤材料Ｓのベルト面画像Ｇを図９のコンピュータ３０に入力することにより，工事現場に搬入される地盤材料Ｓの粒度品質をほぼ連続的に管理することが期待できる。しかし，本発明者らの予備的実験によると，ベルトコンベア１０のベルト面１４と対向する撮像装置５で撮影したベルト面画像Ｇには，地盤材料Ｓだけでなく，地盤材料Ｓ以外の物体（例えば露出したベルト面等）も写り込んでおり，その物体の輪郭を誤って地盤材料Ｓと認識してしまうことが経験された。地盤材料Ｓの誤認識は粒径加積曲線Ｐ（ｄ）の誤差の原因となり，ひいては地盤材料Ｓの粒度品質管理の精度低下につながる。また，地盤材料Ｓを用いる構造物の品質低下を招くことから，地盤材料Ｓの誤認識はできる限り小さく抑えることが必要である。 As described above, for example, the ground material S is placed on the vibrating belt conveyor 10 or the like shown in FIG. By inputting into the computer 30, it is expected that the grain size quality of the ground material S carried into the construction site can be managed almost continuously. However, according to preliminary experiments by the present inventors, the belt surface image G captured by the imaging device 5 facing the belt surface 14 of the belt conveyor 10 contains not only the ground material S but also objects other than the ground material S ( For example, an exposed belt surface, etc.) is also reflected, and it has been experienced that the contour of the object is mistakenly recognized as the ground material S. Misrecognition of the ground material S causes an error in the grain size accumulation curve P(d), which in turn leads to a decrease in the accuracy of the ground material S grain size quality control. In addition, it is necessary to suppress misrecognition of the ground material S as much as possible because it causes deterioration of the quality of the structure using the ground material S.

また，地盤材料Ｓ以外の物体が写り込んだベルト面画像Ｇは，上述した地盤材料Ｓの粒度インデクスＩの誤差の原因ともなりうる。すなわち，粒度インデクスＩは画像Ｇの全体面積Ｅに対する特定粒径ｄ以上の粒状材の面積ｅの総和（Σｅ）の割合Σｅ／Ｅとして定義されるが，全体面積Ｅに地盤材料Ｓ以外の物体が含まれると誤差となり，特定粒径ｄの粒状材の面積ｅに地盤材料Ｓ以外の物体が含まれる場合も誤差となるので，粒度インデクスＩから加積通過率Ｐ（ｄ）を正確に求めることができなくなる。工事現場に搬入される地盤材料Ｓの粒度品質の連続的な管理を実現するためには，画像Ｇに写り込んだ地盤材料Ｓ以外の物体の影響を避ける技術の開発が必要である。 Further, the belt surface image G in which an object other than the ground material S is reflected may cause an error in the particle size index I of the ground material S described above. That is, the grain size index I is defined as the ratio Σe/E of the total sum (Σe) of the area e of granular materials having a specific grain size d or larger to the total area E of the image G. If is included, it will be an error, and if an object other than the ground material S is included in the area e of the granular material with a specific grain size d, it will also be an error. I can't do it. In order to continuously manage the granularity and quality of the ground material S brought into the construction site, it is necessary to develop a technique to avoid the influence of objects other than the ground material S reflected in the image G.

そこで本発明の目的は，ベルトコンベア上の分散した状態の地盤材料の画像から精度よく粒度を判定できる地盤材料の粒度判定方法及びシステムを提供することにある。 SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a method and system for determining the grain size of a ground material that can accurately determine the grain size from an image of the ground material dispersed on a belt conveyor.

一態様において本発明は，図１の実施例及び図２の流れ図並びに図３（Ｂ）及び図４（Ｄ）に示すように，異なる粒径ｄの粒状材が混在した地盤材料Ｓをベルトコンベア１０の帯状ベルト１４に分散した状態で載置し，ベルトコンベア１０と対向する撮像装置５により地盤材料Ｓを載置したベルト面画像Ｇ（図３（Ａ）及び図４（Ａ）参照）を撮影し，ベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを機械学習による推定により検知し，露出ベルト面１４の輪郭領域Ｔ２を地盤材料Ｓ以外の領域としてベルト面画像Ｇから差し引いた除去画像Ｅ（図３（Ｂ）及び図４（Ｄ）参照）を作成し，その除去画像Ｅから地盤材料Ｓの粒度Ｐ（ｄ）を判定してなる地盤材料の粒度判定方法を提供する。 In one embodiment of the present invention, as shown in the embodiment of FIG. 1, the flow chart of FIG. 2, and FIGS . 10 belt-shaped belts 14, and an imaging device 5 facing the belt conveyor 10 captures a belt surface image G (see FIGS. 3(A) and 4(A)) on which the ground material S is placed. Then, the outline area T1 of the ground material S on the belt surface image G and the outline area T2 of the exposed belt surface 14 around the ground material S are detected by estimation by machine learning, and the outline area T2 of the exposed belt surface 14 is detected. A removed image E (see FIGS. 3(B) and 4(D)) is created by subtracting from the belt surface image G as an area other than the ground material S, and the grain size P(d) of the ground material S is calculated from the removed image E. Provided is a method for determining the grain size of a ground material.

他の態様において本発明は，図１，図３（Ｂ）及び図４（Ｄ）の実施例に示すように，異なる粒径ｄの粒状材が混在した地盤材料Ｓを帯状ベルト１４に分散した状態で載置するベルトコンベア１０，ベルトコンベア１０と対向して地盤材料Ｓを載置したベルト面画像Ｇ（図３（Ａ）及び図４（Ａ）参照）を撮影する撮像装置５，ベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を差し引いた除去画像Ｅ（図３（Ｂ）及び図４（Ｄ）参照）を作成する画素選択手段５０，及び除去画像Ｅから地盤材料Ｓの粒度Ｐ（ｄ）を判定する判定手段４０を備え，画素選択手段５０にベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを検知する検知手段５２（図１のブロック図参照）を含め，その検知手段５２を，地盤材料Ｓ中の各粒状材の輪郭形状とベルト面１４上に分散させた地盤材料Ｓ周りの露出ベルト面１４の輪郭形状とを教師データとして用い，ベルト面画像Ｇからセマンティックセグメンテーション手法の機械学習により生成された地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを推定する輪郭推定モデルとし，画素選択手段５０により露出ベルト面１４の輪郭領域Ｔ２を地盤材料Ｓ以外の領域として除去画像Ｅを作成してなる地盤材料の粒度判定システムを提供する。 In another aspect of the present invention, as shown in the embodiments of FIGS. belt conveyor 10 to be placed in a state, an imaging device 5 for capturing a belt surface image G (see FIGS. 3A and 4A) on which the ground material S is placed facing the belt conveyor 10, and the belt surface Pixel selection means 50 for creating a removed image E (see FIGS. 3B and 4D) obtained by subtracting areas T2 and T3 other than the ground material S from the image G, and the grain size of the ground material S from the removed image E A determination means 40 for determining P(d) is provided , and a pixel selection means 50 detects a contour region T1 of the ground material S on the belt surface image G and a contour region T2 of the exposed belt surface 14 around the ground material S. Detecting means 52 (see the block diagram of FIG. 1), including the detection means 52, the contour shape of each granular material in the ground material S and the exposed belt surface 14 around the ground material S dispersed on the belt surface 14 Using the contour shape as teacher data, the contour region T1 of the ground material S generated by machine learning of the semantic segmentation method from the belt surface image G and the contour region T2 of the exposed belt surface 14 around the ground material S are estimated. A ground material granularity determination system is provided in which a removal image E is created by using a contour estimation model and using the pixel selection means 50 to define the contour region T2 of the exposed belt surface 14 as a region other than the ground material S.

更に他の態様において本発明は，図１，図３（Ｃ），図３（Ｄ）及び図４（Ｄ）の実施例に示すように，異なる粒径ｄの粒状材が混在した地盤材料Ｓを帯状ベルト１４に分散した状態で載置するベルトコンベア１０，ベルトコンベア１０と対向して地盤材料Ｓを載置したベルト面画像Ｇ（図３（Ａ）及び図４（Ａ）参照）を撮影する撮像装置５，ベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を差し引いた除去画像Ｅ（図３（Ｂ）及び図４（Ｄ）参照）を作成する画素選択手段５０，及び除去画像Ｅから地盤材料Ｓの粒度Ｐ（ｄ）を判定する判定手段４０を備え，画素選択手段５０にベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを検知する検知手段５２（図１のブロック図参照）と共に露出ベルト面１４の輪郭領域Ｔ２をベルト面１４の帯幅方向に拡大してベルト面全幅にわたる矩形領域Ｔ４，Ｔ５に変形する領域変形手段５４を含め（図３（Ｃ）参照）を含め，画素選択手段５０によりベルト面全幅にわたる矩形領域Ｔ４，Ｔ５を地盤材料以外の領域として除去画像Ｅを作成してなる地盤材料の粒度判定システムを提供する。In still another aspect, the present invention provides a ground material S in which granular materials with different particle sizes d are mixed, as shown in the examples of FIGS. A belt surface image G (see FIGS. 3A and 4A) on which the ground material S is placed facing the belt conveyor 10 and the belt conveyor 10 is photographed. image pickup device 5, pixel selection means 50 for creating a removed image E (see FIGS. 3(B) and 4(D)) obtained by subtracting the areas T2 and T3 other than the ground material S from the belt surface image G, and the removed image A determination means 40 for determining the grain size P(d) of the ground material S from E is provided, and a pixel selection means 50 is provided with a contour area T1 of the ground material S on the belt surface image G and an exposed belt surface 14 around the ground material S. The outline area T2 of the exposed belt surface 14 is enlarged in the band width direction of the belt surface 14 and transformed into rectangular areas T4 and T5 covering the entire width of the belt surface together with a detection means 52 (see block diagram of FIG. 1) for detecting the outline area T2. Including the area transforming means 54 (see FIG. 3(C)), the pixel selecting means 50 makes the rectangular areas T4 and T5 over the entire width of the belt surface the area other than the ground material, and creates the removed image E of the ground material. A granularity determination system is provided.

好ましい実施例では，図４（Ｂ）～（Ｃ）に示すように，画素選択手段５０にベルト面画像Ｇを所要大きさの複数の分割画像ｇ１～ｇ８（図４（Ｂ）参照）に区分けする分割手段５１（図１のブロック図参照）と各分割画像ｇ１～ｇ８を統合する統合手段５３（図１のブロック図参照）とを含め，検知手段５２により分割画像ｇ１～ｇ８毎に地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを推定し（図４（Ｃ）参照），統合手段５３により各分割画像ｇ１～ｇ８の地盤材料Ｓの輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とをそれぞれ統合してベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とを検知する。 In a preferred embodiment, as shown in FIGS. 4B to 4C, the belt surface image G is divided into a plurality of divided images g1 to g8 (see FIG. 4B) by the pixel selection means 50. The ground material for each of the divided images g1 to g8 is detected by the detection means 52, including the dividing means 51 (see the block diagram in FIG. 1) and the integrating means 53 (see the block diagram in FIG. 1) for integrating the divided images g1 to g8. S and the contour region T2 of the exposed belt surface 14 around the ground material S are estimated (see FIG. 4(C)). The outline area T1 of the ground material S on the belt surface image G and the outline area T2 of the exposed belt surface 14 are detected by integrating T1 and the outline area T2 of the exposed belt surface 14 respectively.

本発明による地盤材料の粒度判定方法及びシステムは，異なる粒径ｄの粒状材が混在した地盤材料Ｓをベルトコンベア１０の帯状ベルト１４に分散した状態で載置し，ベルトコンベア１０と対向する撮像装置５により地盤材料Ｓを載置したベルト面画像Ｇを撮影し，ベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓ周りの露出ベルト面１４の輪郭領域Ｔ２とを機械学習による推定により検知し，露出ベルト面１４の輪郭領域Ｔ２を地盤材料Ｓ以外の領域としてベルト面画像Ｇから差し引いた除去画像Ｅを作成し，その除去画像Ｅから地盤材料Ｓの粒度Ｐ（ｄ）を判定するので，次の有利な効果を奏する。 The method and system for determining the grain size of a ground material according to the present invention is to place a ground material S in which granular materials having different particle sizes d are mixed on the belt-shaped belt 14 of the belt conveyor 10 in a dispersed state, The belt surface image G on which the ground material S is placed is photographed by the device 5, and the contour region T1 of the ground material S on the belt surface image G and the contour region T2 of the exposed belt surface 14 around the ground material S are machine-learned. , and the contour area T2 of the exposed belt surface 14 is subtracted from the belt surface image G as an area other than the ground material S to create a removed image E, and from the removed image E, the grain size P (d) of the ground material S is determined, the following advantageous effects are obtained.

（イ）ベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を差し引いた除去画像Ｅを作成し，地盤材料Ｓのみが写り込んだ除去画像Ｅに基づいて粒径加積曲線Ｐ（ｄ）を作成することにより，地盤材料Ｓ以外の物体の影響を避け，精度の高い粒径加積曲線Ｐを作成することができる。
（ロ）色彩や光反射率の相違等を利用してベルト面画像Ｇに写り込んだ地盤材料Ｓ以外の物体を識別検知することもできるが，地盤材料Ｓ以外の物体は地盤材料Ｓ中の粒状材とは通常異なる輪郭形状であることから，輪郭形状を利用してベルト面画像Ｇの地盤材料Ｓ以外の領域Ｔ２，Ｔ３を識別検知することにより，色彩や光反射率の相違等が利用できない条件下でも地盤材料Ｓの粒径加積曲線Ｐを精度よく作成できる。
（ハ）地盤材料Ｓの輪郭形状は一定ではなく，地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３の輪郭形状も一定ではないが，セマンティックセグメンテーション手法の機械学習により生成された輪郭推定モデルを利用することにより，ベルト面画像Ｇから地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を精度よく識別検知し，地盤材料Ｓの粒径加積曲線Ｐの精度を高めることが可能である。判定対象の地盤材料Ｓを教師データとして輪郭推定モデルのパラメタを最適化することにより，判定対象の地盤材料Ｓに応じて地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３の検知精度を高めることもできる。 (a) A removed image E is created by subtracting the areas T2 and T3 other than the ground material S from the belt surface image G, and the grain size accumulation curve P(d) is based on the removed image E in which only the ground material S is reflected. By creating , it is possible to avoid the influence of objects other than the ground material S and create a grain size accumulation curve P with high accuracy.
(B) It is possible to identify and detect objects other than the ground material S reflected in the belt surface image G by using differences in color and light reflectance, etc., but objects other than the ground material S are Since the outline shape is usually different from that of the granular material, the areas T2 and T3 other than the ground material S of the belt surface image G are identified and detected by using the outline shape, so that the difference in color and light reflectance can be used. The grain size accumulation curve P of the ground material S can be created with high accuracy even under conditions where it is not possible.
(C) The contour shape of the ground material S is not constant, and the contour shapes of the contour areas T2 and T3 other than the ground material S are also not constant, but the contour estimation model generated by machine learning using the semantic segmentation method is used. Thus, the contour regions T2 and T3 other than the ground material S can be identified and detected from the belt surface image G with high accuracy, and the accuracy of the grain size addition curve P of the ground material S can be improved. By optimizing the parameters of the contour estimation model using the ground material S to be determined as teaching data, it is possible to increase the detection accuracy of the contour regions T2 and T3 other than the ground material S according to the ground material S to be determined.

以下，添付図面を参照して本発明を実施するための形態及び実施例を説明する。
は，本発明の地盤材料の粒度判定システムの一実施例のブロック図である。は，本発明の粒度判定方法を示す流れ図の一例である。は，本発明で用いるベルト面画像Ｇ及び除去画像Ｅの一例の説明図である。は，本発明で用いる地盤材料の輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とを検知する検知手段の一例の説明図である。は，粒度判定で用いる従来の振動ベルトコンベアの一例の説明図である。は，地盤材料Ｓを用いて構造材料を製造する従来方法の流れ図である。は，地盤材料Ｓの粒径加積曲線ｄの一例を示すグラフである。は，地盤材料Ｓの粒度インデクスＩｉと加積通過率Ｐ（ｄｉ）との変換式Ｋを示すグラフの一例である。は，地盤材料Ｓの撒き出し画像Ｇから粒径加積曲線ｄを作成する従来システムの一例の説明図である。 DETAILED DESCRIPTION OF THE INVENTION Embodiments and examples for carrying out the present invention will be described below with reference to the accompanying drawings.
1 is a block diagram of an embodiment of a ground material grain size determination system according to the present invention; FIG. is an example of a flow chart showing the granularity determination method of the present invention. 3A and 3B are explanatory diagrams of an example of a belt surface image G and a removed image E used in the present invention; FIG. 4 is an explanatory diagram of an example of detection means for detecting the contour region T1 of the ground material and the contour region T2 of the exposed belt surface 14 used in the present invention. is an explanatory diagram of an example of a conventional vibrating belt conveyor used in particle size determination. 1 is a flow diagram of a conventional method for manufacturing structural materials using a foundation material S. FIG. is a graph showing an example of the grain size addition curve d of the ground material S; is an example of a graph showing a conversion formula K between the grain size index Ii of the ground material S and the cumulative passage rate P(di). is an explanatory diagram of an example of a conventional system for creating a grain size accumulation curve d from a bare image G of a ground material S. FIG.

図１は，工事現場付近で調達した地盤材料Ｓを原材料として構造材料（ＣＳＧ，セメント改良土，コンクリート等）を製造する工事現場に適用した本発明の粒度判定システムの実施例を示す。図６に示すように，工事現場に搬入された地盤材料Ｓはホッパー４に投入され，ホッパー４から混合装置９（ミキサー等）に搬送して水Ｗ及びセメントＣを添加することにより構造材料とする。図示例の粒度判定システムは，ホッパー４から混合装置９への搬送路に設けたベルトコンベア１０と，そのベルトコンベア１０上の分散させた状態の地盤材料Ｓを含むベルト面画像Ｇを継続的に撮影する撮像装置５と，そのベルト面画像Ｇを順次入力して地盤材料Ｓの粒径加積曲線Ｐ（ｄ）を作成するコンピュータ３０（画像処理装置）とにより構成されている。 FIG. 1 shows an embodiment of the grain size determination system of the present invention applied to a construction site where structural materials (CSG, cement-improved soil, concrete, etc.) are manufactured using ground material S procured near the construction site as raw material. As shown in Fig. 6, the ground material S brought into the construction site is put into the hopper 4, transported from the hopper 4 to the mixing device 9 (mixer, etc.), and water W and cement C are added to form the structural material. do. The particle size determination system of the illustrated example continuously captures a belt surface image G including the ground material S in a dispersed state on the belt conveyor 10 provided on the conveying path from the hopper 4 to the mixing device 9 and the belt surface image G on the belt conveyor 10. It is composed of an image pickup device 5 for photographing, and a computer 30 (image processing device) for sequentially inputting the belt surface image G and creating a grain size accumulation curve P(d) of the ground material S.

図示例のベルトコンベア１０は，例えば図５（Ａ）を参照して上述した振動ベルトコンベア１０と同様のものとすることができる。すなわち，ベルトコンベア１０の上流部を振動装置２２付き第１支持体２１で支持された振動エリアとし，その下流部を第１支持体２１から縁切りされた第２支持体１７で支持された無振動エリアとし，ベルト面１４を搬送方向に沿って振動エリアと無振動エリアとに区分けしたものとする。ベルト面１４の上流側の振動エリアに地盤材料Ｓを投入・載置し，搬送しながら所要振動数で振動させて分散させ，分散後の地盤材料Ｓを無振動エリアに搬送して無振動状態で観察可能とする。ベルト面１４は，後述する撮像装置５の画像Ｇ中に写り込んだ地盤材料Ｓとベルト面１４とが色彩，光反射率等で識別検知できるように選択又は加工することができる。 The illustrated belt conveyor 10 may be similar to the vibrating belt conveyor 10 described above with reference to FIG. 5A, for example. That is, the upstream part of the belt conveyor 10 is a vibration area supported by a first support 21 with a vibration device 22, and the downstream part is a non-vibration area supported by a second support 17 cut off from the first support 21. It is assumed that the belt surface 14 is divided into a vibrating area and a non-vibrating area along the conveying direction. The ground material S is put and placed in the vibration area on the upstream side of the belt surface 14, and dispersed by being vibrated at a required frequency while being transported. observable with The belt surface 14 can be selected or processed so that the ground material S and the belt surface 14 reflected in the image G of the imaging device 5 to be described later can be identified and detected by color, light reflectance, and the like.

ベルトコンベア１０の振動エリアの長さ（第１支持体２１の搬送方向長さ）及び振動周波数は，粒度の判定に適した地盤材料Ｓの均等な分散が得られるように適宜選択することができる。必要に応じて，図５（Ｂ）に示すように上流側の第１支持体２１と下流側の第２支持体１７との間に両支持体２１，１７から縁切りされた振動装置２４付き第３支持体２３を設け，地盤材料Ｓを二段階で振動させてもよい。図５（Ｂ）のベルトコンベア１０は，振動装置２２，２４を同一又は異なる振動条件で振動させることができ，地盤材料Ｓの性状に応じて振動条件の組み合わせを変化させることもでき，より粒度の判定に適した地盤材料Ｓの均等な分散を得ることが期待できる。振動装置２４付き第３支持体２３は，第１支持体２１と同様に，支持脚によって工事現場の基盤上に支持することができる。 The length of the vibration area of the belt conveyor 10 (the length of the first support 21 in the conveying direction) and the vibration frequency can be appropriately selected so as to obtain a uniform distribution of the ground material S suitable for determining the grain size. . If necessary, as shown in FIG. 5B, between the first support 21 on the upstream side and the second support 17 on the downstream side, a second support 24 with a vibration device 24 cut off from both supports 21 and 17 may be provided. 3 supports 23 may be provided to vibrate the soil material S in two stages. The belt conveyor 10 in FIG. 5(B) can vibrate the vibrating devices 22 and 24 under the same or different vibration conditions, and can change the combination of vibration conditions according to the properties of the ground material S. It can be expected to obtain a uniform distribution of the ground material S suitable for the determination of . Like the first support 21, the third support 23 with the vibration device 24 can be supported on the foundation of the construction site by supporting legs.

ただし，本発明で用いるベルトコンベア１０は振動ベルトコンベアに限定されるわけではなく，地盤材料Ｓを搬送しながら均等に分散させることができる他のベルトコンベアとしてもよい。例えば，地盤材料Ｓを載置するベルト面１４の上流側（上流部分）と対向させてレーキ等の敷き均し板を取り付けたベルトコンベア１０を用い，ベルト面１４上に載置した地盤材料Ｓを搬送しながら敷き均し板によって均等に分散させることも可能である。更に本発明は，ベルトコンベア１０に載置した地盤材料Ｓを搬送しながら分散させるものに限定されず，地盤材料Ｓを適当に分散させたのち又は分散させつつベルトコンベア１０に載置してもよい。例えばホッパー４の投入出口に適当な分散器具を取り付け，ホッパー４から投入される地盤材料Ｓを分散器具で適当に分散した状態でベルトコンベア１０のベルト面１４に載置することもできる。 However, the belt conveyor 10 used in the present invention is not limited to the vibrating belt conveyor, and may be other belt conveyors capable of evenly distributing the ground material S while conveying it. For example, using a belt conveyor 10 equipped with a leveling plate such as a rake facing the upstream side (upstream portion) of the belt surface 14 on which the ground material S is placed, the ground material S placed on the belt surface 14 It is also possible to evenly disperse the particles with a leveling plate while conveying the particles. Furthermore, the present invention is not limited to dispersing while conveying the ground material S placed on the belt conveyor 10, and the ground material S may be placed on the belt conveyor 10 after or while being dispersed appropriately. good. For example, an appropriate dispersing device may be attached to the inlet port of the hopper 4, and the ground material S fed from the hopper 4 may be placed on the belt surface 14 of the belt conveyor 10 in a state of being appropriately dispersed by the distributing device.

図示例の撮像装置５は，ベルトコンベア１０の無振動エリアのベルト面１４と対向するように配置されており，上流部において分散させた状態の地盤材料Ｓを含むベルト面画像Ｇを振動させずに撮影することができる。また，図示例の撮像装置５は，遮光板又は遮光カーテンで覆われた撮影建屋５ａ内に照明装置（フラッシュ等）と共に配置されている。撮像装置５と組み合わせる照明装置は，後述するように画像Ｇ中に写り込んだ地盤材料Ｓと地盤材料Ｓ以外の物体とが色彩，光反射率等で識別検知できるように選択することができる。図示例においてベルトコンベア１０上の地盤材料Ｓは，無振動エリアを搬送されながら撮影建屋５ａの内部に進入し，所要の照明状態に維持された状態でベルト面画像Ｇ中に写り込む。撮像装置５で撮影するベルト面画像Ｇは静止画又は動画の何れとすることも可能である。 The imaging device 5 in the illustrated example is arranged so as to face the belt surface 14 in the non-vibrating area of the belt conveyor 10, and does not vibrate the belt surface image G including the ground material S in a dispersed state in the upstream part. can be photographed. The imaging device 5 in the illustrated example is arranged together with a lighting device (flash, etc.) in a photography building 5a covered with a light shielding plate or a light shielding curtain. The lighting device combined with the imaging device 5 can be selected so that the ground material S reflected in the image G and objects other than the ground material S can be identified and detected by color, light reflectance, etc., as will be described later. In the illustrated example, the ground material S on the belt conveyor 10 enters the inside of the photographing building 5a while being conveyed in the non-vibration area, and is reflected in the belt surface image G while being maintained under the required lighting conditions. The belt surface image G captured by the imaging device 5 can be either a still image or a moving image.

図示例のコンピュータ３０は，図９を参照して上述したコンピュータと同様に，入力装置３１と出力装置３２と記憶手段３５とを有する。記憶手段３５には，ベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を検知するために必要な画素選択パラメタＦ，地盤材料Ｓの粒度インデクスＩｉを加積通過率Ｐ（ｄｉ）に変換する関係式Ｋその他のパラメタを記憶する。また内蔵プログラムとして，撮像装置５からベルト面画像Ｇを入力する入力手段３３と，そのベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を差し引いた除去画像Ｅを作成する画素選択手段５０と，その画素選択手段５０で作成された除去画像Ｅを取り込んで地盤材料Ｓの粒度Ｐ（ｄ）を判定する判定手段４０と，地盤材料Ｓの粒度Ｐ（ｄ）その他の判定結果を出力装置３２に出力する出力手段３４とを有する。 The illustrated computer 30 has an input device 31, an output device 32, and a storage means 35, like the computer described above with reference to FIG. The storage means 35 stores a pixel selection parameter F necessary for detecting regions T2 and T3 other than the ground material S from the belt surface image G, and converts the grain size index Ii of the ground material S into an additive passage rate P(di). Stores the relational expression K and other parameters. Further, as built-in programs, an input means 33 for inputting the belt surface image G from the imaging device 5, and a pixel selection means 50 for creating a removed image E by subtracting areas T2 and T3 other than the ground material S from the belt surface image G. , and a determination means 40 for determining the grain size P(d) of the ground material S by taking in the removed image E created by the pixel selection means 50, and an output device 32 for determining the grain size P(d) of the ground material S and other determination results. and an output means 34 for outputting to.

図示例のコンピュータ３０の判定手段４０は，除去画像Ｅから地盤材料Ｓ中の各粒状材の輪郭を検出する検出手段４１と，各粒状材の輪郭から粒径ｄ及び面積ｅを求めて複数の粒径ｄの粒度インデクスＩ（ｄ）を算出する算出手段４２と，算出した複数の粒径ｄの粒度インデクスＩ（ｄ）を変換式Ｋにより加積通過率Ｐに変換して粒径加積曲線Ｐ（ｄ）を作成する作成手段４３と，作成した粒径加積曲線Ｐ（ｄ）に基づき地盤材料Ｓの粒度品質を判定・評価する評価手段４４とを含んでいる。これらの検出手段４１，算出手段４２，作成手段４３，及び評価手段４４は，判定対象を地盤画像Ｓではなく除去画像Ｅとしている点以外は，図９を参照して上述した特許文献３の粒度計測システムと同様のものとすることできる。 The determination means 40 of the computer 30 of the illustrated example includes a detection means 41 for detecting the contour of each granular material in the ground material S from the removed image E, and a plurality of particle sizes d and area e obtained from the contour of each granular material. Calculation means 42 for calculating the particle size index I(d) of the particle size d, and the calculated particle size index I(d) of the plurality of particle sizes d are converted into the cumulative passage rate P by the conversion formula K, and the particle size is added. It includes a creation means 43 for creating a curve P(d) and an evaluation means 44 for judging and evaluating the grain size quality of the ground material S based on the created grain size accumulation curve P(d). The detection means 41, the calculation means 42, the creation means 43, and the evaluation means 44 use the removed image E instead of the ground image S as the object of determination. It can be similar to the measurement system.

他方，コンピュータ３０の画素選択手段５０は，例えば図４（Ａ）に示すようなベルト面画像Ｇ中に写り込んだ地盤材料Ｓ以外の領域（例えばベルト面１４の露出領域やベルト面１４の外側領域）を検知し，その地盤材料Ｓ以外の領域（例えばベルト面１４の露出領域Ｔ２）をベルト面画像Ｇから差し引いて図４（Ｄ）に示すような除去画像Ｅを作成する。図４（Ａ）は，撮像装置５で撮影したベルト面画像Ｇの一例を示している。ベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２を差し引いた除去画像Ｅ（図４（Ｄ）参照）を作成し，地盤材料Ｓの領域Ｔ１のみが写り込んだ除去画像Ｅを判定手段４０に入力して地盤材料Ｓの粒度Ｐ（ｄ）を判定することにより，地盤材料Ｓ以外の物体の写り込みによる影響を避け，地盤材料Ｓの粒度Ｐを精度よく判定することができる。 On the other hand, the pixel selection means 50 of the computer 30 selects areas other than the ground material S reflected in the belt surface image G as shown in FIG. area) is detected, and the area other than the ground material S (for example, the exposed area T2 of the belt surface 14) is subtracted from the belt surface image G to create a removed image E as shown in FIG. 4(D). FIG. 4A shows an example of the belt surface image G photographed by the imaging device 5. FIG. A removed image E (see FIG. 4(D)) is created by subtracting an area T2 other than the ground material S from the belt surface image G, and the removed image E in which only the area T1 of the ground material S is reflected is input to the determination means 40. By determining the grain size P(d) of the ground material S by doing so, the grain size P of the ground material S can be accurately determined while avoiding the influence of reflection of objects other than the ground material S.

図３（Ａ）は，ベルト面画像Ｇに写り込んだ地盤材料Ｓ以外の物体の輪郭領域Ｔ２，Ｔ３の一例を示している。ベルトコンベア１０上に載置した地盤材料Ｓは，ベルト面１４に分散された状態で載置されるが，ベルト面１４の全体を覆うように分散するとは限らず，地盤材料Ｓの周りにベルト面１４の露出した領域Ｔ２が残ることがある（図４（Ａ）も参照）。画素選択手段５０は，例えば地盤材料Ｓとベルト面１４との色彩や光反射率の相違に基づいて，ベルト面画像Ｇ中に写り込んだ地盤材料Ｓの輪郭領域Ｔ１とその地盤材料Ｓの周りの露出ベルト面１４の輪郭領域Ｔ２とを検知する検知手段５２を含んでいる。画素選択手段５０は，検知手段５２で検知された露出ベルト面１４の輪郭領域Ｔ２を地盤材料Ｓ以外の領域としてベルト面画像Ｇから差し引くことにより，図３（Ｂ）のような除去画像Ｅを作成することができる。 FIG. 3A shows an example of contour areas T2 and T3 of objects other than the ground material S reflected in the belt surface image G. FIG. The ground material S placed on the belt conveyor 10 is placed in a dispersed state on the belt surface 14. However, the ground material S is not necessarily dispersed so as to cover the entire belt surface 14. An exposed region T2 of surface 14 may remain (see also FIG. 4A). The pixel selection means 50 selects the outline region T1 of the ground material S reflected in the belt surface image G and the periphery of the ground material S based on the difference in color and light reflectance between the ground material S and the belt surface 14, for example. and sensing means 52 for sensing the contour area T2 of the exposed belt surface 14 of the belt. The pixel selection means 50 subtracts the contour area T2 of the exposed belt surface 14 detected by the detection means 52 from the belt surface image G as an area other than the ground material S, thereby obtaining a removed image E as shown in FIG. can be created.

また図３（Ａ）に示すように，ベルト面画像Ｇにはベルト面１４の外側（帯幅方向の片側又は両側）に，例えばベルトコンベア１０を設置した現場の基盤が写り込むことがある。画素選択手段５０に含まれる検知手段５２は，例えばベルト面１４とその外側（例えば現場の基盤）との色彩や光反射率の相違に基づいて，ベルト面画像Ｇ中に写り込んだベルト面１４の外側の輪郭領域Ｔ３とを検知する（ベルト面１４とその外側との境界を検知する）ことができる。画素選択手段５０は，検知手段５２で検知された露出ベルト面１４の輪郭領域Ｔ２と共にベルト面外側の輪郭領域Ｔ３を地盤材料Ｓ以外の領域としてベルト面画像Ｇから差し引いくことにより，図３（Ｂ）のような除去画像Ｅを作成することができる。 Further, as shown in FIG. 3A, the belt surface image G may include, for example, the site base on which the belt conveyor 10 is installed, on the outside of the belt surface 14 (on one side or both sides in the width direction). The detection means 52 included in the pixel selection means 50 detects the belt surface 14 reflected in the belt surface image G based on the difference in color or light reflectance between the belt surface 14 and the outside thereof (for example, the substrate on site). (detecting the boundary between the belt surface 14 and its outside). The pixel selection means 50 subtracts the outline area T2 of the exposed belt surface 14 detected by the detection means 52 and the outline area T3 outside the belt surface from the belt surface image G as areas other than the ground material S, thereby obtaining the image shown in FIG. A removal image E such as B) can be created.

ベルト面画像Ｇに写り込んだ地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３は，地盤材料Ｓ中の粒状材の輪郭とは異なる輪郭形状となるのが通常であることから，地盤材料Ｓとの色彩や光反射率の相違を利用できない条件下であっても，地盤材料Ｓ中の粒状材の輪郭形状との相違を利用して地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を検知することができる。粒状材の輪郭形状は一定ではなく，地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３の輪郭形状も一定ではないが，必ずしも輪郭形状が一定でない物体の画像（例えば手書き文字の画像）から機械学習により物体の属するクラス及び輪郭形状（物体と背景との境界）を予測する技術（セマンティックセグメンテーション手法）が開発されている（非特許文献１参照）。このようなセマンティックセグメンテーション手法の機械学習により生成された輪郭推定モデルを，ベルト面画像Ｇから地盤材料Ｓの輪郭領域Ｔ１と地盤材料Ｓ以外の物体の輪郭形状Ｔ２，Ｔ３とを識別検知する画素選択手段５０の検知手段５２とすることができる。 The contour regions T2 and T3 other than the ground material S reflected in the belt surface image G usually have a contour shape different from the contour of the granular material in the ground material S. Even under conditions in which differences in light reflectance cannot be used, contour regions T2 and T3 other than the ground material S can be detected by utilizing the difference from the contour shape of the granular material in the ground material S. The contour shape of the granular material is not constant, and the contour shapes of the contour regions T2 and T3 other than the ground material S are also not constant. A technique (semantic segmentation method) for predicting the class to which the object belongs and the contour shape (the boundary between the object and the background) has been developed (see Non-Patent Document 1). A contour estimation model generated by machine learning of such a semantic segmentation method is used for pixel selection for identifying and detecting the contour region T1 of the ground material S and the contour shapes T2 and T3 of objects other than the ground material S from the belt surface image G. It can be the sensing means 52 of the means 50 .

セマンティックセグメンテーション手法の機械学習によって画像中の物体のクラス及び輪郭形状（物体と背景との境界）を推定する検知手段５２の一例は，畳み込み層とブーリング層とを積み重ねた畳み込みニューラルネットワーク（ＣＮＮ）プログラムにより構成することができる。セマンティックセグメンテーション手法は，物体の属するクラスと共に画像中の物体が存在する領域を推定する物体検出手法の一種であり，通常の物体検出手法では物体を囲む矩形領域が推定されるのに対し，セマンティックセグメンテーション手法では画素中の画素レベルまで物体の領域と背景の領域とを切り分け，物体の輪郭形状（物体と背景の境界）を推定することができる（非特許文献１参照）。 An example of the detection means 52 that estimates the class and contour shape (boundary between the object and the background) of the object in the image by machine learning of the semantic segmentation method is a convolutional neural network (CNN) program that stacks convolution layers and Boolean layers. It can be configured by The semantic segmentation method is a type of object detection method that estimates the area in the image where the object exists along with the class to which the object belongs. In this method, the object area and the background area are separated down to the pixel level in pixels, and the contour shape of the object (the boundary between the object and the background) can be estimated (see Non-Patent Document 1).

例えば，図４（Ａ）のようなベルト面画像Ｇに基づき，予め露出ベルト面１４の輪郭領域Ｔ２（及び，必要な場合はベルト面外側の輪郭領域Ｔ３）を多角形で近似した複数の教師データを調製し，その教師データを画素選択手段５０の検知手段５２に入力して地盤材料Ｓ以外の輪郭領域Ｔ２を推定させ，その推定結果と教師データとの誤差が最小となるように画素選択パラメターＦ（畳み込みニューラルネットワークの重みデータ等のパラメタ）を最適化する。最適化されたパラメタＦを用いて検知手段５２の畳み込みニューラルネットワークを構成することにより，教師データ以外の通常のベルト面画像Ｇから地盤材料Ｓの輪郭領域Ｔ１と地盤材料Ｓ以外の輪郭領域Ｔ２を精度よく識別検知することができる。 For example, based on the belt surface image G shown in FIG. 4(A), a plurality of teachers are obtained by approximating the outline area T2 of the exposed belt surface 14 (and the outline area T3 outside the belt surface if necessary) with polygons. Data is prepared, and the teacher data is input to the detection means 52 of the pixel selection means 50 to estimate the contour region T2 other than the ground material S, and pixel selection is performed so that the error between the estimation result and the teacher data is minimized. Optimize the parameter F (parameter such as the weight data of the convolutional neural network). By configuring the convolutional neural network of the detection means 52 using the optimized parameter F, the contour region T1 of the ground material S and the contour region T2 other than the ground material S are detected from the normal belt surface image G other than the teacher data. It can be identified and detected with high accuracy.

地盤材料Ｓの輪郭領域Ｔ１，及び地盤材料Ｓ以外の輪郭領域Ｔ２は，判定対象の地盤材料Ｓの性状に応じて相違することがある。例えば，粘土質の土を多く含むコア材等の地盤材料Ｓと，礫や砕石を多く含むロック材等の地盤材料Ｓとは，分散させたときの輪郭形状が相違しうる。従って，上述した画素選択パラメターＦを最適化するための教師データは，判定対象の地盤材料Ｓに基づいて調整することが望ましい。また，作業現場毎に地盤材料Ｓ中の粒状材の大きさ，色，凹凸，サイズ，形状等が変わってくるので，画素選択パラメターＦを最適化するための教師データは作業現場毎に調整することが望ましい。判定対象の地盤材料Ｓに応じた教師データによって画素選択パラメターＦを最適化することにより，その地盤材料Ｓに応じた輪郭領域Ｔ１と地盤材料Ｓ以外の輪郭領域Ｔ２との識別検知精度を高めることができる。 The contour area T1 of the ground material S and the contour area T2 other than the ground material S may differ depending on the property of the ground material S to be determined. For example, a ground material S such as a core material containing a large amount of clayey soil and a ground material S such as a rock material containing a large amount of gravel or crushed stone may differ in outline shape when dispersed. Therefore, it is desirable to adjust the training data for optimizing the pixel selection parameter F described above based on the ground material S to be determined. In addition, since the size, color, unevenness, size, shape, etc. of the granular materials in the ground material S change for each work site, the training data for optimizing the pixel selection parameter F is adjusted for each work site. is desirable. By optimizing the pixel selection parameter F using teacher data corresponding to the ground material S to be determined, the accuracy of discrimination detection between the contour area T1 corresponding to the ground material S and the contour area T2 other than the ground material S is improved. can be done.

好ましくは，図１の画素選択手段５０に示すように，ベルト面画像Ｇを所要大きさの複数の分割画像ｇ１～ｇ８に区分けする分割手段５１と，複数の分割画像ｇ１～ｇ８を統合する統合手段５３とを画素選択手段５０に含める。画素選択手段５０の検知手段５２により，例えば図４（Ａ）のような撮像装置５から出力されるベルト面の帯幅全体にわたるベルト面画像Ｇから直接的に地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を検知することもできるが，全体の画像Ｇが大きくなると輪郭領域Ｔ２，Ｔ３の輪郭形状が複雑になり，複数の教師データの相互間のバラツキが大きくなるため，画素選択パラメターＦを最適化しても，地盤材料Ｓの輪郭領域Ｔ１と地盤材料Ｓ以外の輪郭領域Ｔ２とを識別検知する精度の低下が経験された。 Preferably, as shown in pixel selection means 50 in FIG. 1, dividing means 51 for dividing the belt surface image G into a plurality of divided images g1 to g8 of a required size, and integration for integrating the plurality of divided images g1 to g8. means 53 are included in the pixel selection means 50; The detection means 52 of the pixel selection means 50 directly detects the contour area T2 other than the ground material S from the belt surface image G covering the entire band width of the belt surface output from the imaging device 5 as shown in FIG. T3 can also be detected, but as the entire image G becomes larger, the contour shapes of the contour regions T2 and T3 become more complicated, and the variation among the plurality of teacher data increases. However, a decrease in the accuracy of identifying and detecting the contour region T1 of the ground material S and the contour region T2 other than the ground material S was experienced.

例えば，図４（Ａ）のベルト面画像Ｇを，分割手段５１によって図４（Ｂ）のように所要大きさの複数の分割画像（例えば一辺３００～２４００画素程度の矩形画像）ｇ１～ｇ８に区分し，検知手段５２によって図４（Ｃ）のように分割画像ｇ１～ｇ８毎に地盤材料Ｓの輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とを推定する。そのうえで，統合手段５３によって図４（Ｄ）のように各分割画像ｇ１～ｇ８の地盤材料Ｓの輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とをそれぞれ統合することにより，ベルト面画像Ｇ上の地盤材料Ｓの輪郭領域Ｔ１と露出ベルト面１４の輪郭領域Ｔ２とを検知する。ベルト面画像Ｇを複数の分割画像ｇ１～ｇ８に分割したうえで地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を推定することにより，地盤材料Ｓの輪郭領域Ｔ１と地盤材料Ｓ以外の輪郭領域Ｔ２との識別検知精度の向上を図ることができる。 For example, the belt surface image G in FIG. 4(A) is divided into a plurality of divided images (for example, rectangular images of about 300 to 2400 pixels on a side) g1 to g8 of a required size as shown in FIG. 4(B) by the dividing means 51. Then, the detection means 52 estimates the outline area T1 of the ground material S and the outline area T2 of the exposed belt surface 14 for each of the divided images g1 to g8 as shown in FIG. 4(C). After that, as shown in FIG. A contour region T1 of the ground material S and a contour region T2 of the exposed belt surface 14 are detected. By dividing the belt surface image G into a plurality of divided images g1 to g8 and estimating the contour regions T2 and T3 other than the ground material S, the contour region T1 of the ground material S and the contour region T2 other than the ground material S are obtained. identification detection accuracy can be improved.

図２は，図１のシステムを用いて地盤材料Ｓの粒度Ｐ（ｄ）を判定する方法の流れ図を示す。以下，図２の流れ図を参照して図１のシステムを説明する。図２のステップＳ１０１は，コンピュータ１０の関係式設定手段４６により，図８を参照して上述した地盤材料Ｓの粒度インデクスＩと加積通過率Ｐ（ｄ）との変換式Ｋを設定する初期処理を示す。例えば，工事現場付近で調達した地盤材料Ｓについて篩い分けにより複数の粒径ｄの加積通過率Ｐ（ｄ）をそれぞれ求め，他方でその地盤材料Ｓのベルト面画像Ｇから各粒径ｄの粒度インデクスＩ（ｄ）を算出し，それらから図８のような変換式Ｋ（例えば回帰モデル）を求めてコンピュータ３０の記憶手段３５に設定・記憶する。ただし，関係式Ｋは予め記憶手段３５に記憶されていれば足り，予め記憶されている場合は図２のステップＳ１０１は省略可能であり，パラメタ設定手段５６は本発明に必須のものではない。 FIG. 2 shows a flow diagram of a method for determining the grain size P(d) of a subsurface material S using the system of FIG. The system of FIG. 1 will now be described with reference to the flow diagram of FIG. Step S101 in FIG. 2 is an initial stage for setting the conversion formula K between the grain size index I of the ground material S and the cumulative passage rate P(d) described above with reference to FIG. Indicates processing. For example, for the ground material S procured near the construction site, the cumulative passage rate P(d) of a plurality of particle sizes d is obtained by sieving. A granularity index I(d) is calculated, a conversion formula K (for example, a regression model) as shown in FIG. However, it suffices if the relational expression K is stored in advance in the storage means 35, and if it is stored in advance, step S101 in FIG. 2 can be omitted, and the parameter setting means 56 is not essential to the present invention.

また図２のステップＳ１０２は，コンピュータ１０のパラメタ設定手段５６により，画素選択手段５０の検知手段５２に教師データを入力して，最適化された画素選択パラメターＦを設定する初期処理を示す。例えば，検知手段５２をセマンティックセグメンテーション手法の機械学習によって画像Ｇ中の物体の輪郭形状を推定する輪郭推定モデルとし，工事現場付近で調達した地盤材料Ｓのベルト面画Ｇに基づき，予め露出ベルト面１４の輪郭領域Ｔ２（及び，必要な場合はベルト面外側の輪郭領域Ｔ３）を多角形で近似した複数の教師データを調製する。そして，その教師データと検知手段５２の推定結果との誤差が最小となるように画素選択パラメターＦを最適化し，最適化した画素選択パラメターＦをコンピュータ３０の記憶手段３５に設定・記憶する。ただし，画素選択パラメターＦも予め記憶手段３５に記憶されていれば足り，予め記憶されている場合は図２のステップＳ１０２は省略可能であり，パラメタ設定手段５６も本発明に必須のものではない。 Step S102 in FIG. 2 shows initial processing for inputting teacher data to the detection means 52 of the pixel selection means 50 by the parameter setting means 56 of the computer 10 and setting the optimized pixel selection parameter F. FIG. For example, the detection means 52 is a contour estimation model that estimates the contour shape of the object in the image G by machine learning of the semantic segmentation method, and based on the belt surface image G of the ground material S procured near the construction site, the belt surface exposed in advance. A plurality of teaching data are prepared by approximating the 14 contour areas T2 (and the contour areas T3 outside the belt surface if necessary) with polygons. Then, the pixel selection parameter F is optimized so that the error between the teacher data and the estimation result of the detection means 52 is minimized, and the optimized pixel selection parameter F is set and stored in the storage means 35 of the computer 30 . However, it is sufficient if the pixel selection parameter F is stored in advance in the storage means 35, and if it is stored in advance, step S102 in FIG. 2 can be omitted, and the parameter setting means 56 is not essential to the present invention. .

図２のステップＳ１０３において，画像装置５によりベルトコンベア１０上の分散させた状態の地盤材料Ｓを含むベルト面画像Ｇを撮影し，そのベルト面画像Ｇをコンピュータ３０の入力手段３３に入力する。ステップＳ１０４において，入力したベルト面画像Ｇを画素分離手段５０に送り，画素分離手段５０の検知手段５２においてベルト面画像Ｇ中に写り込んだ地盤材料Ｓ以外の物体の輪郭領域Ｔ２，Ｔ３を検知し，その領域Ｔ２，Ｔ３をベルト面画像Ｇから差し引いて除去画像Ｅを作成する。必要に応じて，図４を参照して上述したように，画素分離手段５０の検知手段５２の前段，後段にそれぞれ分割手段５１，統合手段５３を設けてもよい。分割手段５１によりベルト面画像Ｇを複数の矩形分割画像ｇ１～ｇ８に区分し，検知手段５２により分割画像ｇ１～ｇ８毎に地盤材料Ｓ以外の物体の輪郭領域Ｔ２，Ｔ３を検知したのち，統合手段５３により各分割画像ｇ１～ｇ８を統合することによりベルト面画像Ｇ上の地盤材料Ｓ以外の物体の輪郭領域Ｔ２，Ｔ３を検知し，その輪郭領域Ｔ２，Ｔ３をベルト面画像Ｇから差し引いて除去画像Ｅを作成する。 In step S103 of FIG. 2, the belt surface image G including the ground material S in the dispersed state on the belt conveyor 10 is photographed by the image device 5, and the belt surface image G is input to the input means 33 of the computer 30. In step S104, the input belt surface image G is sent to the pixel separation means 50, and the detection means 52 of the pixel separation means 50 detects contour regions T2 and T3 of objects other than the ground material S reflected in the belt surface image G. Then, a removed image E is created by subtracting the regions T2 and T3 from the belt surface image G. FIG. If necessary, as described above with reference to FIG. 4, the division means 51 and integration means 53 may be provided in the front stage and the rear stage of the detection means 52 of the pixel separation means 50, respectively. The dividing means 51 divides the belt surface image G into a plurality of rectangular divided images g1 to g8, and the detection means 52 detects contour areas T2 and T3 of objects other than the ground material S for each of the divided images g1 to g8, and then integrates them. By integrating the divided images g1 to g8 by means 53, the contour areas T2 and T3 of the object other than the ground material S on the belt surface image G are detected, and the contour areas T2 and T3 are subtracted from the belt surface image G. A removed image E is created.

図２のステップＳ１０５において，画素分離手段５０の作成した除去画像Ｅを判定手段４０に送り，判定手段４０の検出手段４１により除去画像Ｅ中の個々の粒状材の輪郭を検出する。上述した画素分離手段５０において個々の粒状材の輪郭形状が既に検知済みである場合は，検出手段４１による検出を省略し，画素分離手段５０で検出した輪郭を用いてもよい。次いでステップＳ１０６において，算出手段４２により各粒状材の粒径ｄｉ及び面積ｅを求め，複数の粒径ｄの粒度インデクスＩ（ｄ）を算出する。更にステップＳ１０７において，複数の粒径ｄｉの粒度インデクスＩｉを作成手段４３に入力し，作成手段４３において記憶手段３５の変換式Ｋにより各粒径ｄｉの粒度インデクスＩｉを加積通過率Ｐ（ｄｉ）に変換して粒径加積曲線Ｐ（ｄ）を作成する。 In step S105 of FIG. 2, the removed image E created by the pixel separation means 50 is sent to the determination means 40, and the detection means 41 of the determination means 40 detects the outline of each granular material in the removed image E. FIG. When the contour shape of each granular material has already been detected by the pixel separation means 50 described above, the detection by the detection means 41 may be omitted and the contour detected by the pixel separation means 50 may be used. Next, in step S106, the particle size di and area e of each granular material are determined by the calculating means 42, and the particle size index I(d) of the plurality of particle sizes d is calculated. Further, in step S107, the particle size index Ii of a plurality of particle sizes di is input to the creating means 43, and the particle size index Ii of each particle size di is calculated by the conversion formula K of the storage means 35 in the creating means 43. ) to create the particle size addition curve P(d).

図２のステップＳ１０８は，ステップＳ１０７で作成した粒状材料Ｓの粒径加積曲線Ｐ（ｄ）を評価手段４４に入力し，評価手段４４において地盤材料Ｓの粒度品質を判定・評価する処理を示す。例えば，地盤材料Ｓの粒度規定範囲を示す最粗粒標本Ｔｒの粒径加積曲線Ｐｒ（ｄ）と最細粒標本Ｔｓの粒径加積曲線Ｐｓ（ｄ）とを予めコンピュータ３０の記憶手段３５にしておき，評価手段４４により粒径加積曲線Ｐｒ，Ｐｓと比較することにより粒状材料Ｓの粒径加積曲線Ｐ（ｄ）が規定範囲内であるか否かを評価する（図７を参照）。規定範囲外であると評価された場合は，例えばコンピュータ３０の出力手段３４を介してディスプレイ・プリンタ等の出力装置３２に警報を出力し，必要に応じて粒状材料Ｓの粒度を調整することができる。 In step S108 in FIG. 2, the grain size accumulation curve P(d) of the granular material S created in step S107 is input to the evaluation means 44, and the evaluation means 44 performs processing for judging and evaluating the grain size quality of the ground material S. show. For example, the grain size accumulation curve Pr(d) of the coarsest-grained sample Tr and the grain size accumulation curve Ps(d) of the finest-grained sample Ts, which indicate the specified range of grain size of the ground material S, are stored in advance in the storage means of the computer 30. 35, the evaluation means 44 compares the particle size addition curves Pr and Ps to evaluate whether or not the particle size addition curve P(d) of the granular material S is within a specified range (Fig. 7 ). If it is evaluated as being out of the specified range, for example, an alarm is output to the output device 32 such as a display printer via the output means 34 of the computer 30, and the particle size of the granular material S can be adjusted as necessary. can.

図２のステップＳ１０８において粒状材料Ｓの粒度品質が規定範囲内であると判定された場合は，例えば粒径加積曲線Ｐ（ｄ）を記憶手段３５に累積記憶したのち，ステップＳ１０９において粒状材料Ｓの粒度判定を継続するか否かを判断する。継続する場合はステップＳ１０３へ戻り，次回に入力されたベル面画像Ｇに基づいて粒径加積曲線Ｐを作成して記録する。粒径加積曲線Ｐを累積記憶しておくことにより，次回以降のステップＳ１０８において，浄化手段４４により粒状材料Ｓの粒度の経時的変化（粒度変動）を評価することも可能である。 If it is determined in step S108 in FIG. 2 that the particle size quality of the granular material S is within the specified range, for example, after cumulatively storing the particle size accumulation curve P(d) in the storage means 35, in step S109 the granular material It is determined whether or not to continue the granularity determination of S. When continuing, the process returns to step S103, and the grain size accumulation curve P is created and recorded based on the bell surface image G input next time. By cumulatively storing the particle size accumulation curve P, it is also possible to evaluate the temporal change (particle size fluctuation) of the particle size of the granular material S by the purifying means 44 in step S108 after the next time.

本発明は，粒状材料Ｓを分散させたベルト面画像Ｇから地盤材料Ｓ以外の領域Ｔ２，Ｔ３を差し引いた除去画像Ｅを作成し，地盤材料Ｓのみが写り込んだ除去画像Ｅに基づいて粒径加積曲線Ｐ（ｄ）を作成するので，地盤材料Ｓ以外の物体の影響を避けつつ，精度の高い粒径加積曲線Ｐを作成することができる。また，色彩や光反射率の相違等を利用してベルト面画像Ｇに写り込んだ地盤材料Ｓ以外の物体を識別検知することもできるが，輪郭形状を利用してベルト面画像Ｇの地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を識別検知することにより，色彩や光反射率の相違等が利用できない条件下でも地盤材料Ｓの粒径加積曲線Ｐを精度よく作成することができる。 In the present invention, a removed image E is created by subtracting regions T2 and T3 other than the ground material S from a belt surface image G in which the granular material S is dispersed, and based on the removed image E in which only the ground material S is reflected, grains Since the diameter addition curve P(d) is created, the grain size addition curve P can be created with high accuracy while avoiding the influence of objects other than the ground material S. In addition, it is possible to identify and detect an object other than the ground material S reflected in the belt surface image G by using the difference in color and light reflectance, etc. By identifying and detecting the contour areas T2 and T3 other than S, the grain size accumulation curve P of the ground material S can be accurately created even under conditions where differences in color, light reflectance, etc. cannot be used.

こうして本発明の目的である「ベルトコンベア上の分散した状態の地盤材料の画像から精度よく粒度を判定できる地盤材料の粒度判定方法及びシステム」を提供することができる。 In this way, it is possible to provide a ground material grain size determination method and system capable of accurately determining the grain size from an image of ground material dispersed on a belt conveyor, which is the object of the present invention.

図２のステップ１０４では，例えば図３（Ａ）のようなベルト面画像Ｇ中に写り込んだ地盤材料Ｓ以外の輪郭領域Ｔ２，Ｔ３を検知し，その領域Ｔ２，Ｔ３をベルト面画像Ｇから差し引いて図３（Ｂ）のような除去画像Ｅを作成できるが，本発明者の予備的実験によると，地盤材料Ｓ以外の輪郭領域Ｔ２の識別検知に漏れが発生しうることも経験された。この場合，識別検知が漏れた地盤材料Ｓ以外の輪郭領域Ｔ２は，識別検知された他の輪郭領域Ｔ２とベルト面１４の帯幅方向に並んでいることが多いことを見出した。このようにベルト面１４の帯幅方向に並んだ複数の輪郭領域Ｔ２の一部分に識別検知の漏れが生じた原因の詳細は不明であるが，図１に示すようにベルト面１４の帯幅方向は搬送方向と直角に交差しており，ベルト面１４に同時に投入・載置された地盤材料Ｓはほぼ帯幅方向に並ぶことが一原因であると推定される。 In step 104 in FIG. 2, for example, outline regions T2 and T3 other than the ground material S reflected in the belt surface image G as shown in FIG. Although it is possible to create a removed image E as shown in FIG. . In this case, it was found that the contour region T2 other than the ground material S for which identification detection was omitted is often aligned with other contour regions T2 identified and detected in the band width direction of the belt surface 14 . Although the details of the cause of the omission of identification detection in a part of the plurality of contour regions T2 aligned in the width direction of the belt surface 14 are unknown, as shown in FIG. intersects the conveying direction at a right angle, and it is presumed that one of the reasons is that the ground materials S that have been fed and placed on the belt surface 14 at the same time are aligned substantially in the band width direction.

図３（Ｃ）は，上述した識別検知の漏れた地盤材料Ｓ以外の輪郭領域Ｔ２の影響を避けるため，ベルト面画像Ｇ中の識別検知された地盤材料Ｓ以外の輪郭領域Ｔ２をベルト面１４の帯幅方向に拡大し，ベルト面全幅にわたる矩形領域Ｔ４，Ｔ５に変形する実施例を示している。例えば，図１に示すように，識別検知された輪郭領域Ｔ２をベルト面全幅にわたる矩形領域Ｔ４，Ｔ５に変形する領域変形手段５４を画素選択手段５０に含めることができる。 In FIG. 3C, in order to avoid the influence of the contour region T2 other than the ground material S for which identification detection has failed, the contour region T2 other than the ground material S identified and detected in the belt surface image G is replaced with the belt surface 14 is expanded in the belt width direction and deformed into rectangular areas T4 and T5 covering the entire width of the belt surface. For example, as shown in FIG. 1, the pixel selecting means 50 may include an area transforming means 54 for transforming the identified and detected outline area T2 into rectangular areas T4 and T5 extending over the entire width of the belt surface.

領域変形手段５４によってベルト面画像Ｇ中の識別検知された地盤材料Ｓ以外の輪郭領域Ｔ２をベルト面全幅にわたる矩形領域Ｔ４，Ｔ５に変更したのち，画素選択手段５０によってベルト面全幅にわたる矩形領域Ｔ４，Ｔ５をベルト面画像Ｇから差し引いて図３（Ｄ）のような除去画像Ｅを作成する。図３（Ｄ）に示すように，ベルト面画像Ｇから矩形領域Ｔ４，Ｔ５を差し引くと，ベルト面全幅が地盤材料Ｓのみからなる矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃが残るので，例えば残された矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃを結合して単一の除去画像Ｅを作成することができる。或いは，残された矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃからそれぞれ除去画像Ｅを作成することもできる。 After the area transforming means 54 changes the contour area T2 other than the ground material S identified and detected in the belt surface image G into rectangular areas T4 and T5 covering the entire width of the belt surface, the pixel selecting means 50 converts the rectangular area T4 covering the entire width of the belt surface. , T5 are subtracted from the belt surface image G to create a removed image E as shown in FIG. 3(D). As shown in FIG. 3(D), when the rectangular regions T4 and T5 are subtracted from the belt surface image G, rectangular regions T1a, T1b, and T1c whose entire width of the belt surface is composed only of the ground material S remain. A single ablation image E can be created by combining the regions T1a, T1b, T1c. Alternatively, the removed images E can be created from the remaining rectangular areas T1a, T1b, and T1c.

更に，図３（Ｄ）のような除去画像Ｅを画素選択手段５０から判定手段４０に送って粒度インデクスＩ（ｄ）を算出する。例えば，図３（Ｄ）において複数の矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃを結合して単一の除去画像Ｅとした場合は，判定手段４０の算出手段４２において単一の除去画像Ｅから粒度インデクスＩ（ｄ）を算出する。また，図３（Ｄ）において複数の矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃをそれぞれ除去画像Ｅとした場合は，判定手段４０の算出手段４２において各矩形領域Ｔ１ａ，Ｔ１ｂ，Ｔ１ｃの粒度インデクスを算出し，それらの平均値を除去画像Ｅの粒度インデクスＩ（ｄ）とすることもできる。 Further, the removed image E as shown in FIG. 3(D) is sent from the pixel selection means 50 to the determination means 40 to calculate the granularity index I(d). For example, in FIG. 3D, when a plurality of rectangular areas T1a, T1b, and T1c are combined to form a single removed image E, the granularity index I Calculate (d). Further, when a plurality of rectangular areas T1a, T1b, and T1c in FIG. 3D are each set as the removed image E, the calculation means 42 of the determination means 40 calculates the granularity index of each rectangular area T1a, T1b, and T1c, The average value thereof can also be used as the granularity index I(d) of the removed image E. FIG.

図３（Ｄ）のようにベルト面全幅にわたる矩形領域Ｔ４，Ｔ５を差し引いた除去画像Ｅの粒度インデクスＩ（ｄ）から，判定手段４０の作成手段４３において粒径加積曲線Ｐ（ｄ）を作成することにより，上述した識別検知の漏れが発生しうる帯幅方向に並んだ地盤材料Ｓ以外の輪郭領域Ｔ２の影響を避けることができ，精度の高い粒径加積曲線Ｐを作成することができる。 From the particle size index I(d) of the removed image E obtained by subtracting the rectangular areas T4 and T5 over the entire width of the belt surface as shown in FIG. By creating it, it is possible to avoid the influence of the contour area T2 other than the ground material S arranged in the band width direction, which may cause the leakage of identification detection described above, and to create a grain size accumulation curve P with high accuracy. can be done.

１…採取場（地山）１ａ…破砕装置
２…ストックヤード３…運搬装置（トラック等）
４…ホッパー５…撮像装置
５ａ…撮影建屋６…分離装置
７ａ…重量計測器７ｂ…含水率計測器
８ａ…セメント供給装置８ｂ…水供給装置
１０…ベルトコンベア１１…駆動プーリ（ドライブプーリ）
１２…駆動装置１３…テールプーリ
１４…帯状ベルト１５，１６…キャリアローラ
１７…第２支持体１８…スナッププーリ
２１…第１支持体２２…振動装置
２３…第３支持体２４…振動装置
２５，２６…振動制御装置２７…駆動制御装置
３０…コンピュータ３１…入力装置
３２…出力装置３３…入力手段
３４…出力手段３５…記憶手段
４０…判定手段
４１…検出手段４２…算出手段
４３…作成手段４３ａ…調整手段
４３ｂ…合成手段４４…評価手段
４５…演算手段４６…関係式設定手段
４７…推定手段
５０…画素選択手段５１…画像分割手段
５２…検知手段５３…画像統合手段
５４…領域変形手段５６…パラメタ設定手段
Ｇ…ベルト面画像Ｅ…除去画像
Ｓ…地盤材料Ｋ…変換式
Ｔ１…地盤材料の輪郭領域Ｔ２…露出ベルト面の輪郭領域
Ｔ３…ベルト面の外側領域Ｔ４，Ｔ５…変形後の矩形領域
Ｐ…加積通過率，粒径加積曲線
Ｕ，Ｒ…（微小粒状材の加積通過率推定用の）関数
Ｆ…画素選択用パラメタ 1... Collection site (ground) 1a... Crushing device 2... Stockyard 3... Transport device (truck, etc.)
4... Hopper 5... Imaging device 5a... Shooting building 6... Separating device 7a... Weight measuring instrument 7b... Moisture content measuring instrument 8a... Cement supply device 8b... Water supply device 10... Belt conveyor 11... Drive pulley (drive pulley)
DESCRIPTION OF SYMBOLS 12... Driving device 13... Tail pulley 14... Strip belt 15, 16... Carrier roller 17... Second support 18... Snap pulley 21... First support 22... Vibration device 23... Third support 24... Vibration device 25, 26 Vibration control device 27 Drive control device 30 Computer 31 Input device 32 Output device 33 Input means 34 Output means 35 Storage means 40 Determination means 41 Detection means 42 Calculation means 43 Creation means 43a Adjustment means 43b Synthesis means 44 Evaluation means 45 Calculation means 46 Relational expression setting means 47 Estimation means 50 Pixel selection means 51 Image division means 52 Detection means 53 Image integration means 54 Area deformation means 56 Parameter setting means G Belt surface image E Removed image S Ground material K Conversion formula T1 Outline area of ground material T2 Outline area of exposed belt surface T3 Outer area of belt surface T4, T5 Rectangle after deformation Area P: Accumulated passage rate, grain size accumulation curve U, R: Function (for estimating accumulated passage rate of fine particulate material) F: Parameter for pixel selection

Claims

A ground material containing a mixture of granular materials with different particle sizes is placed in a dispersed state on a belt-shaped belt of a belt conveyor, and an image of the belt surface on which the ground material is placed is taken by an imaging device facing the belt conveyor. The outline area of the ground material on the surface image and the outline area of the exposed belt surface around the ground material are detected by estimation by machine learning, and the outline area of the exposed belt surface is subtracted from the belt surface image as the area other than the ground material. A method for determining the grain size of a ground material, which comprises creating a removed image and determining the grain size of the ground material from the removed image.

A belt conveyor that places ground materials containing mixed granular materials of different particle sizes in a dispersed state on a belt-shaped belt, an imaging device that captures an image of the belt surface on which the ground materials are placed facing the belt conveyor, and the belt surface. A pixel selection means for creating a removed image by subtracting an area other than the ground material from the image, and a determination means for determining the grain size of the ground material from the removed image , wherein the pixel selection means is provided with the ground material on the belt surface image. The detection means, including detection means for detecting the contour area and the contour area of the exposed belt surface around the said foundation material, shall be adapted to the contour shape of each particulate material in the said foundation material and the ground material distributed over the belt surface. Using the contour shape of the surrounding exposed belt surface as training data, the contour area of the ground material generated by machine learning using the semantic segmentation method from the belt surface image and the contour area of the exposed belt surface around the ground material are estimated. a grain size determination system for a ground material, in which a removed image is created by using the pixel selection means to define the contour area of the exposed belt surface as an area other than the ground material.

A belt conveyor that places ground materials containing mixed granular materials of different particle sizes in a dispersed state on a belt-shaped belt, an imaging device that captures an image of the belt surface on which the ground materials are placed facing the belt conveyor, and the belt surface. A pixel selection means for creating a removed image by subtracting an area other than the ground material from the image, and a determination means for determining the grain size of the ground material from the removed image , wherein the pixel selection means is provided with the ground material on the belt surface image. detection means for detecting the outline area and the outline area of the exposed belt surface around the ground material; A ground material grain size determination system in which a rectangular area covering the entire width of the belt surface is used as an area other than the ground material by the pixel selection means to create a removed image .

4. The system according to claim 2 , wherein said pixel selection means includes area transforming means for enlarging said outline area of said exposed belt surface in the band direction of said belt surface and transforming it into a rectangular area covering the entire width of said belt surface. A ground material grain size determination system in which a rectangular area covering the entire width of the belt surface is used as an area other than the ground material by a pixel selection means to create a removed image.