A Coal Distribution Method Based on Video Recognition

Abstract

This invention provides a coal quantity distribution method based on video recognition, belonging to the field of coal quantity information processing, and includes the following steps: S1: Install cameras capable of coal quantity recognition on each sub-belt to output the instantaneous coal quantity of the corresponding sub-belt; S2: Collect instantaneous coal quantity data from the cameras on each sub-belt via a server; S3: Collect instantaneous speed information of each sub-belt and the main belt via a server; S4: Calculate the coal quantity distribution of each sub-belt using a coal quantity conversion formula; S5: As the instantaneous speed increases, the instantaneous coal quantity accumulates, and the displacement change is calculated based on the coal quantity distribution information of the associated sub-belts; S6: Calculate the coal quantity distribution curve through fitting and smoothing prediction correction, wherein the smoothing prediction correction adopts an exponential smoothing model.

Description

Technical Field

[0001] This invention relates to a method for processing coal quantity information, and more particularly to a method for coal quantity distribution based on video recognition. Background Technology

[0002] Belt conveyor systems have always been a crucial component of coal mine production. To ensure safe production, production units install cameras at fixed points on the belts to monitor coal quantity and belt operation. However, due to the presence of multiple overlapping belts in mines, currently, only the coal quantity at the current location can be observed in real-time via cameras; data fusion is not yet possible. Summary of the Invention

[0003] This invention provides a coal quantity distribution method based on video recognition, which solves the problem of achieving coal quantity distribution through video recognition using a camera. The technical solution is as follows:

[0004] A method for coal quantity distribution based on video recognition includes the following steps:

[0005] S1: Install cameras capable of coal quantity identification on each sub-belt to output the instantaneous coal quantity of the corresponding sub-belt.

[0006] S2: Collect instantaneous coal quantity data from cameras on each sub-belt via the server;

[0007] S3: Collect instantaneous speed information of each sub-belt and main belt through the server;

[0008] S4: The coal quantity distribution of each sub-belt is calculated by fitting the coal quantity conversion formula;

[0009] S5: As the instantaneous speed increases, the instantaneous coal quantity accumulates continuously. The coal quantity distribution information of the associated sub-belt is calculated to change the displacement.

[0010] S6: The coal quantity distribution curve is calculated through the fitting and smoothing prediction correction in step S4. The smoothing prediction correction adopts the exponential smoothing model.

[0011] Furthermore, cameras are installed at designated locations on each sub-belt.

[0012] Furthermore, the designated location is 100 meters away from the main belt overlap point.

[0013] Furthermore, the instantaneous coal quantity of the sub-belt is distributed in the instantaneous coal quantity distribution of the main belt as follows: (instantaneous coal quantity of the sub-belt) * (speed ratio of the main belt to the sub-belt) * (width ratio of the main belt to the sub-belt) * coal quantity conversion coefficient.

[0014] Furthermore, define each t0, t1....tn At time , the speeds of the sub-belt A are respectively AV t0 AV t1 ...AV tn The moving positions of the sub-belt A are AD respectively. t0 AD t1 , ......., AD tn ,but:

[0015] AD t1 =(t1-t0)AV t1 ;

[0016] AD t2 =(t2-t1)AV t2 ; ......

[0017] AD tn =(t n -t n-1 )AV tn .

[0018] Define each t0, t1, ..., t n At time t, the speeds of the sub-belt C are respectively CV. t0 CV t1 ...CV tn The moving positions of the sub-belt A are CD respectively. t0 CD t1 , ......., CD tn ,but:

[0019] CD t1 = (t1-t0)CV t1 ;

[0020] CD t2 =(t2-t1)CV t2 ; ......

[0021] CD tn =(t n -t n-1 )CV tn .

[0022] Similarly, if there are other sub-belts, they should be handled in the same way as described above.

[0023] Furthermore, cameras are installed on each sub-belt at a distance of 100 meters from the main belt. Coal distribution calculations are performed every T meters. The fitted coal distribution for the main belt FD is then: (0, FD-A-Qt1) + (0, FD-C-Qt1), (T, FD-A-Qt2) + (T, FD-C-Qt2), ..., (100, FD-A-Qt2). n )+(100,FD-C-Qt) n );

[0024] Wherein, FD-A-Qtn and FD-C-Qtn represent the coal quantity distribution of sub-belts A and C on the main belt FD, respectively, and t i The time period is represented by i, which is a natural number, and T represents the set distance, ranging from 0.5 to 2.5.

[0025] Similarly, when there are multiple sub-belts, the fitted coal quantity distribution of the main belt FD is the sum of the coal quantity distributions calculated from the multiple sub-belts.

[0026] Furthermore, in step S4, the weighted average of the observations obtained by the exponential smoothing method will decrease exponentially as the observations age.

[0027] The video recognition-based coal quantity distribution method uses video to identify the coal quantity at each detection point and then uses a coal quantity fusion algorithm to fit the coal quantity situation at each position of each conveyor belt. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overlap between the main belt and the sub-belt;

[0029] Figure 2 This is a schematic diagram of the coal quantity distribution on the main conveyor belt;

[0030] Figure 3 This is a schematic diagram of the coal quantity distribution of each sub-belt on the main belt;

[0031] Figure 4 This is a schematic diagram of coal quantity prediction using time series.

[0032] Figure 5 This is a flowchart illustrating the video recognition-based coal quantity distribution method. Detailed Implementation

[0033] The coal quantity distribution method based on video recognition provided by this invention applies an overlap relationship model comprising one main conveyor belt and N sub-conveyor belts. The output ends of the N sub-conveyor belts are connected to the main conveyor belt. Each sub-conveyor belt is equipped with a smart camera at a predetermined location to collect real-time coal quantity data. Using the real-time coal quantity and the sub-conveyor belt speeds, the dynamic coal quantity distribution from the camera installation location on the sub-conveyor belt to the overlap point on the main conveyor belt can be calculated. By combining the speeds of multiple sub-conveyor belts, the identified coal quantity, and the speed of the main conveyor belt with a coal quantity conversion formula, the coal quantity distribution on the main conveyor belt can be calculated and fitted.

[0034] In one embodiment, Figure 1 Regarding the overlap relationship between the main belt and the sub-belts, there are two sub-belts that overlap with the main belt FD: a first sub-belt A and a second sub-belt C. A first camera is installed at point B of the first sub-belt A, and the overlap position between the first sub-belt A and the main belt FD is the first overlap point P1. The length from the first camera to the main belt FD is BP1. A second camera is installed at point E of the second sub-belt C, and the overlap position between the second sub-belt C and the main belt FD is the second overlap point P2. The length from the second camera to the main belt FD is EP2. Figure 2 As shown, the focus is usually on the coal quantity distribution of the main belt, but the coal quantity distribution of each sub-belt cannot be seen. This invention takes into account that the speeds of the first sub-belt A and the second sub-belt C are different from those of the main belt FD. The data stream is fitted using the arithmetic progression method, and all data in the data stream is processed through the coal quantity conversion formula.

[0035] Taking the first sub-belt A as an example: Let the length of the first sub-belt A be A. L The width of the first sub-belt A is A. W The coefficient of the coal quantity conversion formula is A. F ;

[0036] Define each t0, t1, ..., t n At time t, the speeds of the first sub-belt A are respectively AV t0 AV t1 ...AV tn The moving positions of the first sub-belt A are AD. t0 AD t1 , ......., AD tn ,but:

[0037] AD t1 =(t1-t0)AV t1 ;

[0038] AD t2 =(t2-t1)AV t2 ; ......

[0039] AD tn =(t n -t n-1 )AV tn .

[0040] Correspondingly, the instantaneous coal quantity of the first sub-belt A is AQ. t0 AQ t1 ,.......,AQ tn n is a natural number.

[0041] Let the length of the second sub-belt C be C. L Width is C W The coefficient of the coal quantity conversion formula is C. F Define the values ​​for t0, t1, ..., t2. n At time t, the speeds of the second sub-belt C are respectively CV t0 CV t1 , ......, CV tn The moving positions of the second sub-belt C are CD respectively. t0 CD t1 , ......., CD tn ,but:

[0042] CD t1 = (t1-t0)CV t1 ;

[0043] CD t2 =(t2-t1)CV t2 ; ......

[0044] CD tn =(t n -t n-1 )CV tn .

[0045] Correspondingly, the instantaneous coal quantity of the second sub-belt C is CQ. t0 CQ t1 , ......, CQ tn .

[0046] Set the speed of the main belt FD to FD-V. t0 FD-V t1 ,.......,FD-V tn The width of the main belt FD W .

[0047] Suppose a certain t n Time AD tn>BP1 length, at this time the instantaneous coal quantity AQ of the first sub-belt A tn It has already landed on the main belt FD, AQ tn The coal quantity conversion formula is needed, which is the instantaneous coal quantity distribution FD-AQ. tn ;

[0048] Coal quantity conversion formula: FD-AQ tn =(AQ tn )*(FD-V tn ) / (AV tn )*FD W / A W *A F That is, the instantaneous coal quantity distribution of the first sub-belt A on the main belt FD is: (instantaneous coal quantity of the first sub-belt A) * (speed ratio of the main belt FD to the first sub-belt A) * (width ratio of the main belt FD to the first sub-belt A) * coal quantity conversion coefficient.

[0049] Similarly, by analogy, suppose a certain t n Time AD tn >EP2 length, at which point the instantaneous coal quantity CQ of the second sub-belt C. tn It has already landed on the main belt FD, CQ tn The required coal quantity conversion formula is FD-CQ. tn :

[0050] FD-CQ tn =(CQ tn )*(FD-V tn ) / (CV) tn )*FD W / C W *C F That is, the instantaneous coal quantity distribution of the second sub-belt C on the main belt FD is: (instantaneous coal quantity of the second sub-belt C) * (speed ratio of the main belt FD to the second sub-belt C) * (width ratio of the main belt FD to the second sub-belt C) * coal quantity conversion coefficient.

[0051] Among them, FD-AD t0 FD-AD t1 , ...... , FD-AD tn Given the instantaneous coal quantity distribution of the first sub-belt A at various times falling on the instantaneous coal quantity movement position of the main belt FD (the position after the main belt is segmented), we can obtain:

[0052] FD-AD t1 =(t1-t0)FD-V t1 ;

[0053] FD-ADt2 =(t2-t1)FD-F t2 ; ......

[0054] FD-AD tn =(tn-t n-1 FD-V tn .

[0055] Same as the first sub-belt A, FD-CD t0 FD-CD t1 , ...... , FD-CD tn Given the instantaneous coal quantity distribution of the second sub-belt C at various times falling on the instantaneous coal quantity movement position of the main belt FD (the position after the main belt is segmented), we can obtain:

[0056] FD-CD t1 =(t1-t0)FD-V t1 ;

[0057] FD-CD t2 =(t2-t1)FD-V t2 ; ......

[0058] FD-CD tn =(t n -t n-1 FD-V tn .

[0059] like Figure 3 As shown, the main conveyor belt FD is divided into 100-meter segments, with each segment length, for example, 2 meters. The coal quantity moves from the camera installation position on the sub-belt to the main conveyor belt FD as the sub-belt segments move. The calculation can then be performed from FD-A-Qt1 to FD-A-Qt. n FD-C-Qt1 to FD-C-Qt n The respective locations (location index, coal quantity).

[0060] For example: Qt n : Coal quantity at time n, Qt n-m : Coal quantity at time m before time n

[0061] (0, FD-A-Qt) n (2, FD-A-Qt) n-1 ), ... (100, FD-A-Qt) n-m )

[0062] (0, FD-C-Qt) n (2, FD-C-Qt) n-1),...(100, FD-C-Qt) n-m )

[0063] Where 0 represents the position on the belt, FD-A-Qt n and FD-C-Qt n This represents the coal quantity distribution, where m represents a natural number. If there are gaps in the location index, a piecewise displacement fitting method is used to fill the coal quantity distribution using the arithmetic progression method.

[0064] Therefore, the fitted coal quantity distribution for the main conveyor belt FD is: (0, FD-A-Qt1) + (0, FD-C-Qt1), ...,

[0065] (100, FD-A-Qt) n )+(100,FD-C-Qt) n ).

[0066] like Figure 5 As shown, a coal quantity distribution method based on video recognition includes the following steps:

[0067] S1: Install cameras capable of coal quantity identification on each sub-belt to output the instantaneous coal quantity of the corresponding sub-belt.

[0068] S2: Collect instantaneous coal quantity data from cameras on each sub-belt via the server;

[0069] S3: Collect instantaneous speed information of each sub-belt and main belt through the server;

[0070] S4: The coal quantity distribution of each sub-belt is calculated by fitting the coal quantity conversion formula;

[0071] S5: As the instantaneous velocity and instantaneous coal quantity accumulate, calculate FD-A-Dt. n FD-C-Dt n The information is shifted and changed.

[0072] Where the main belt speed is 0, FD-A-Dt n FD-C-Dt n No change.

[0073] S6: Calculation of coal quantity distribution curve: fitting formula + smoothing prediction correction, wherein the smoothing prediction correction adopts the exponential smoothing method model.

[0074] Exponential smoothing is a weighted average of observations, with weights decaying exponentially as observations age. In other words, the more recent the observation, the higher its weight. It can quickly generate reliable forecasts and is applicable to a wide range of time series. Simple exponential smoothing: This method is suitable for forecasting univariate time series data without a clear trend or seasonal pattern. Simple exponential smoothing models the next time step as an exponentially weighted linear function of the observations at previous time steps. It requires a parameter called alpha(a), also known as the smoothing factor or smoothing coefficient, which controls the rate at which the influence of observations at previous time steps decays exponentially, i.e., the rate at which the weights decrease. a is typically set to a value between 0 and 1. A larger value means the model focuses primarily on recent past observations, while a smaller value means more historical data is considered when making forecasts.

[0075] Using an additive model, such as Figure 4 As shown, it is assumed that the trend component ut and the seasonal component st of the time series {xt} are additive, i.e., ideally xt = ut + st, where ut increases (or decreases) linearly with time, and st is the seasonal component of period T. In reality, due to the non-stationarity of the series {xt}, the linear increase rate of its trend component ut and the seasonal component st are only relatively fixed in the short term, while they can change slowly in the long term. Furthermore, xt may contain irregular noise components. Therefore, exponential smoothing (EMA) is needed to continuously calibrate the ut and st components in the model based on the actual observed values ​​xt.

[0076] ut = α * (xt – st-T) + (1 – α) * (ut-1 + vt-1)

[0077] vt = β * (ut – ut-1) + (1 – β) * vt-1

[0078] st = γ * (xt – ut) + (1 – γ) * st-T

[0079] The above three equations contain three smoothing parameters α, β, and γ, all between 0 and 1, which represent the balancing weights between the model's predicted values ​​and the measured back-calculated values. Here, vt represents the linear increasing rate of the trend component ut. Larger parameters α, β, and γ indicate stronger non-stationarity of the time series {xt}, resulting in a shorter predictable timeframe and requiring faster adjustments to the model's components. Conversely, smaller parameters α, β, and γ that match historical data indicate a better model-data fit and a longer predictable timeframe. When the historical data is exhausted and the model transitions from training to prediction, α=β=γ=0, as there is no more data to correct the model. The predicted value of xt is then calculated using the ideal formula xt = ut + st. To determine reasonable parameters α, β, and γ and the predictable timeframe, cross-validation can be used. The historical data is divided into two segments: the first segment is used to train the model, and after its use, the model enters the prediction phase. The resulting predicted values ​​are then compared with the second segment of historical data.

[0080] By fitting historical coal quantity data and predicted values ​​using the least squares method, smooth prediction corrections can be achieved. The following is a diagram illustrating the specific calculation process for the predicted values:

[0081] Given n data points: (x 1, y1),(x 2, y2)...(x n, y n )

[0082] We need to perform curve fitting on these n data points. Observation shows that it approximates a parabola.

[0083] Assume the equation of the curve is in the form: y = a²x 2 +a1x+a 0, Where a0, a1, and a2 are unknown, if we take

[0084] (x 1, Substituting y1 into the equation, we get y1 = a2x1 2 +a1x1+a0, then transform:

[0085]

[0086] Similarly (x) i ),(y i ), i=1,2...n, we can get

[0087] Therefore, it can be combined into the form of evidence:

[0088] Assumption: For A, For T, For x

[0089] Ax=T

[0090] =>A T Ax=A T T

[0091] =>(A T A) -1 A T Ax=(A T A) -1 A T T

[0092] =>x=(A T A) -1 A T T

[0093] The video recognition-based coal quantity distribution method uses video to identify the coal quantity at each detection point and then uses a coal quantity fusion algorithm to fit the coal quantity situation at each position of each conveyor belt.

Claims

1. A coal quantity distribution method based on video recognition, comprising a main conveyor belt and N sub-conveyor belts, wherein the output ends of the N sub-conveyor belts are connected to the main conveyor belt, comprising the following steps: S1: Install cameras capable of coal quantity identification on each sub-belt to output the instantaneous coal quantity of the corresponding sub-belt. S2: Collect instantaneous coal quantity data from cameras on each sub-belt via the server; S3: Collect instantaneous speed information of each sub-belt and main belt through the server; S4: The coal quantity distribution of each sub-belt is calculated by fitting the coal quantity conversion formula; the instantaneous coal quantity distribution of the sub-belt in the main belt is: (instantaneous coal quantity of the sub-belt) * (speed ratio of the main belt to the sub-belt) * (width ratio of the main belt to the sub-belt) * coal quantity conversion coefficient; S5: As the instantaneous speed increases, the instantaneous coal quantity accumulates continuously. The coal quantity distribution information of the associated sub-belt is calculated to change the displacement. S6: The coal quantity distribution curve is calculated through the fitting and smoothing prediction correction in step S4. The smoothing prediction correction adopts the exponential smoothing model. The weighted average of the observations in the exponential smoothing method decays exponentially as the observations age. The trend component ut and the seasonal component st of the time series {xt} are additive. The ut and st components in the model are continuously calibrated based on the actual observed values ​​xt. st is the seasonal component of the period T. ut= α * (xt– st-T) + (1 – α) * (ut-1+ vt-1); vt= β * (ut– ut-1) + (1 – β) * vt-1; st= γ * (xt– ut) + (1 – γ) * st-T; The above three equations contain three smoothing parameters α, β, and γ, all between 0 and 1. These parameters represent the balancing weights between the model's predicted values ​​and the measured inverse values. Here, vt represents the linear increasing rate of the trend component ut.

2. The coal quantity distribution method based on video recognition according to claim 1, characterized in that: Each sub-belt has a camera installed at a designated location.

3. The coal quantity distribution method based on video recognition according to claim 2, characterized in that: The designated location is 100 meters away from the main belt overlap point.

4. The coal quantity distribution method based on video recognition according to claim 1, characterized in that: Define each t0, t1, ..., t n At time , the speeds of the sub-belt A are respectively AV t0 AV t1 ...AV tn The moving positions of the sub-belt A are AD respectively. t0 AD t1 AD tn ,but: AD t1 =(t1-t0)AV t1 ; A-D t2 =(t2-t1)A-V t2 ; ...... A-D tn =(t n -t n-1 )A-V tn 。 5. The coal quantity distribution method based on video recognition according to claim 4, characterized in that: Define each t0, t1, ..., t n At time t, the speeds of the sub-belt C are respectively CV. t0 CV t1 ...CV tn The moving positions of the sub-belt A are CD respectively. t0 CD t1 ,.......,CD tn ,but: C-D t1 =(t1-t0)C-V t1 ; C-D t2 =(t2-t1)C-V t2 ; ...... C-D tn =(t n -t n-1 )C-V tn 。 6. The coal quantity distribution method based on video recognition according to claim 5, characterized in that: Sub-belts A and C are equipped with cameras at a distance of 100 meters from the main belt. Coal distribution calculations are performed every T meters. The fitted coal distribution for the main belt FD is then: (0, FD-A-Qt1) + (0, FD-C-Qt1), (T, FD-A-Qt2) + (T, FD-C-Qt2), ..., (100, FD-A-Qt2). n )+(100,FD-C-Qt) n ); Wherein, FD-A-Qtn and FD-C-Qtn represent the coal quantity distribution of sub-belts A and C on the main belt FD, respectively, and t i The time period is represented by i, which is a natural number, and T represents the set distance, ranging from 0.5 to 2.5.

Images