A fruit moving target detection method for visual appearance detection

Abstract

The application discloses a fruit motion target detection method for visual appearance detection, and establishes a fruit motion target detection target function by detecting the first several frames in a video of the fruit motion target, obtains weight coefficient recommended values, middle value recommended values and sidelobe recommended values of a triangular component of the fruit motion target detection, and performs adaptive background modeling of the triangular component, so that the weight coefficient recommended values, the middle value recommended values and the sidelobe recommended values of the triangular component of the fruit motion target detection are used in the triangular component probability density function corresponding to each pixel point in the fruit motion target video in the visual appearance detection for matching detection.

Description

Technical Field

[0001] This invention belongs to the field of fruit and vegetable quality detection technology. Specifically, it involves establishing a fruit moving target detection objective function in the first few frames of a video that detects moving targets of fruit, and obtaining recommended values ​​of weight coefficients, median values, and side lobes of the triangular components for fruit moving target detection. These values ​​are then used for matching and detection of pixels in subsequent frames, thereby completing the detection of moving targets of fruit. Background Technology

[0002] With the development of computer vision algorithms and the improvement of hardware platform processing speed, visual appearance inspection technology is being used more and more widely in fruit sorting systems. Among the many visual appearance inspection technologies, moving target detection is of great significance and is one of the important foundations for external quality inspection. Moving target detection is mainly based on background subtraction, but background subtraction cannot be applied to scenes with changing lighting. It is evident that these problems necessitate better background modeling methods to solve them. Therefore, many adaptive background modeling methods have been proposed to address the numerous problems in motion detection under conditions of slowly changing backgrounds. Summary of the Invention

[0003] The purpose of this invention is to establish a moving target detection objective function for fruit by detecting the first few frames of a video of moving targets in fruit, and to obtain the recommended values ​​of the weight coefficients, median values, and sidelobe values ​​of the triangular components for moving target detection in fruit. This allows for adaptive background modeling of the triangular components, thereby achieving the task of moving target detection when visually inspecting fruit.

[0004] According to the design scheme provided by the present invention, a method for detecting moving targets in fruit for visual appearance inspection includes the following steps:

[0005] Step 1: Construct the probability density function of the triangular component corresponding to a certain pixel in the video for detecting moving fruit targets in visual appearance detection, using several triangular components for detecting moving fruit targets, including the weight coefficients, median values, and side lobes of the triangular components for detecting moving fruit targets, as well as the gray value at a certain pixel at a certain time, as independent variables.

[0006] Step 2: Establish the objective function for detecting moving fruit targets by detecting the first few frames of the video, and introduce the Lagrange daily number to ensure that the weight coefficients of the triangular components for detecting moving fruit targets meet the normalization condition.

[0007] Step 3: After differentiating the objective function for fruit moving target detection with respect to the weight coefficients of the triangular components of fruit moving target detection to zero, obtain the weight coefficients of the triangular components of fruit moving target detection that make the objective function for fruit moving target detection have a minimum value;

[0008] Step 4: Obtain the intermediate value of the triangular components of the fruit moving target detection when the derivative of the objective function of fruit moving target detection with respect to the intermediate value of the triangular components of fruit moving target detection is zero;

[0009] Step 5: Obtain the sidelobe of the fruit moving target detection triangular component when the objective function for fruit moving target detection has a minimum value by taking the derivative of the objective function with respect to the sidelobe of the fruit moving target detection triangular component as zero.

[0010] Step 6: By introducing specific calculation formulas for the weight coefficients, median values, and side lobes of the fruit motion target detection triangular components that satisfy the objective function of fruit motion target detection, iterate repeatedly until the sum of the differences between the weight coefficients of the current round and the previous round of fruit motion target detection triangular components is less than or equal to the triangular component weight coefficient difference threshold, thereby outputting the recommended values ​​for the weight coefficients, median values, and side lobes of the fruit motion target detection triangular components;

[0011] Step 7: In visual appearance detection, the recommended values ​​of the weight coefficients, median values, and sidelobe values ​​of the triangular components corresponding to each pixel in the video of fruit movement are used for matching detection in the probability density function of the triangular components of fruit movement detection.

[0012] Furthermore, step 1 specifically includes:

[0013] In visual appearance inspection, the grayscale value x of a video at pixel (m,n) and time t is set to detect the moving target of fruit. m,n The probability density function of the triangular component corresponding to (t) is y m,n (t), the specific calculation formula is as follows:

[0014]

[0015] Where k is the index of the triangulation component for fruit moving target detection, Ξ is the set of indexes for the triangulation component for fruit moving target detection, and λ k Let θ be the weighting coefficient of the triangular component for detecting the k-th moving fruit target. k ε is the median value of the triangular components for detecting the k-th moving fruit target. k The sidelobe of the triangular component for detecting the k-th moving fruit target is u, which is the step function, and m and n are the x and y coordinates of the pixel, respectively.

[0016] Furthermore, step 2 specifically includes:

[0017] To solve for the weight coefficient λ of the triangular component for detecting the k-th moving fruit object. k , median θ k and side lobe ε k Let's assume that the first T frames of the video used to detect moving fruit targets are used to establish a Lagrange daily number γ to achieve λ. k The objective function min{F} for fruit moving target detection that satisfies the normalization condition is calculated using the following formula:

[0018]

[0019] Among them, setting Θ is the time set constructed from the previous T frames;

[0020] Furthermore, step 3 specifically includes:

[0021] By the F to the λ k Differentiation yields the following specific calculation formula:

[0022]

[0023] By the F to the λ k There exists an extremum such that min{F} exists, which shows that... Therefore, we can conclude that:

[0024]

[0025] Considering After sorting, we can obtain:

[0026]

[0027] Therefore, we can obtain λ that satisfies min{F}. k The specific calculation formula is as follows:

[0028]

[0029] Furthermore, step 4 specifically includes:

[0030] By the F to the θ k Differentiation yields The specific calculation formula is as follows:

[0031]

[0032] Among them, to simplify the representation, let

[0033] Considering and Therefore Approximately:

[0034]

[0035] By the F to the θ k There exists an extremum such that min{F} exists, which shows that... Therefore, we can conclude that:

[0036]

[0037] make as well as Therefore, we can conclude that:

[0038]

[0039] After sorting, we can obtain:

[0040]

[0041] make as well as Therefore, the aforementioned θ k The specific calculation formula for a quintic unary equation with variable A is as follows:

[0042]

[0043] According to the Tianheng formula, we can set...

[0044] Considering that the values ​​of H2 and H3 are small, they can be approximated as 0 for ease of calculation. Then we can obtain θ that satisfies min{F} under this condition. k The specific calculation formula is as follows:

[0045]

[0046] Where max is the function that takes the maximum value, and abs is the function that takes the positive value.

[0047] Furthermore, step 5 specifically includes:

[0048] By the F to the ε k The opposite number σ k Differentiation yields The specific calculation formula is as follows:

[0049]

[0050] For simplification, we can set...

[0051] After simplification, we can approximate the following:

[0052]

[0053] By the F to the σ k and ε k There exists an extremum such that min{F} exists, which shows that... and Therefore, we can conclude that:

[0054]

[0055] make as well as By x m,n (t) and y m,n (t) are all greater than 0, and 0 ≤ y m,n Since (t)≤1, we know that D4, D2, and D0 are all greater than 0. Therefore, we can obtain the aforementioned θ. k The specific calculation formula for a quartic equation with one variable is as follows:

[0056]

[0057] It can be known Then, from experience, we can obtain that ε exists that satisfies min{F}. k The specific calculation formula is as follows:

[0058]

[0059] Furthermore, the specific steps in step 6 are as follows:

[0060] Step 6-1: Set the iteration number h to 0, and statistically analyze the grayscale values ​​of different pixels in the first T frames of the video detecting the moving fruit target. Take the average value of similar grayscale values ​​from highest to lowest frequency and set it as θ. k (h=0), and set θ k The number of occurrences of similar gray values ​​divided by num(Θ)·num(Ω) is set as λ. k (h=0), where num is a function to calculate the number of elements in the set, by using θ k (h=0), by satisfying the existence of ε in min{F} k The specific calculation formula for ε is given. k (h=0);

[0061] Step 6-2: Replace h with h+1, and change λ k (h-1) and ε kSubstituting (h-1) into θ that satisfies min{F} k The specific calculation formula for θ is given. k (h);

[0062] Step 6-3: Place λ k (h-1) and θ k (h) Substitute ε that satisfies min{F} k The specific calculation formula for ε is given. k (h);

[0063] Step 6-4: ε k (h) and θ k (h) Substitute λ that satisfies min{F} k The specific calculation formula for λ is given. k (h);

[0064] Step 6-5: Place λ k (h) and λ k Compare (h-1), if Then proceed to step 6-6; otherwise proceed to step 6-2, where δ is the threshold for the difference in the weight coefficients of the triangular components.

[0065] Step 6-6: Place λ k (h), θ k (h) and ε k (h) are recommended weighting coefficients for the triangular components of fruit moving target detection. Median Recommended Value and sidelobe recommendation value Output the results.

[0066] Furthermore, step 7 specifically includes:

[0067] When detecting moving targets in fruit for visual appearance inspection, for a pixel (i,j) within frame l of a video, it is recommended to first use the triangulation component k′ used for that pixel in frame l-1. Perform a judgment to match and detect, if If the match is found, it is considered a successful match; otherwise, a recommendation value is made based on different weighting coefficients. Matching and detection are performed sequentially from largest to smallest. If a certain k in the Ξ exists, it can make... If true, then pixel (i,j) uses the k-th fruit moving target detection triangulation component. Attached Figure Description

[0068] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0069] Figure 1This is a flowchart of a fruit moving target detection method for visual appearance inspection in an embodiment of the present invention;

[0070] Figure 2 This is the iterative process in this embodiment of the invention for outputting the recommended values ​​of the weight coefficients, median values, and sidelobe values ​​of the triangular components for detecting moving targets in fruits;

[0071] Figure 3 It is a grayscale image frame in the video in this embodiment of the invention;

[0072] Figure 4 This describes the differences in the weight coefficients of the triangular components during different iterations in the embodiments of the present invention. Detailed Implementation

[0073] To gain a deeper understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0074] The embodiments of the present invention provide a method for detecting moving targets in fruit for visual appearance inspection.

[0075] Please refer to Figure 1 , Figure 1 This is a flowchart of a fruit moving target detection method for visual appearance inspection according to an embodiment of the present invention, which specifically includes the following steps:

[0076] Step 1: Construct the probability density function of the triangular components corresponding to a pixel in the video for detecting moving fruit targets in visual appearance detection, using several triangular components for detecting moving fruit targets, including the weight coefficients, median values, and side lobes of the triangular components, as well as the grayscale value of a pixel at a certain time, as independent variables. The specific method for determining this function is as follows:

[0077] In visual appearance inspection, the grayscale value x of a video at pixel (m,n) and time t is set to detect the moving target of fruit. m,n The probability density function of the triangular component corresponding to (t) is y m,n (t), the specific calculation formula is as follows:

[0078]

[0079] Where k is the index of the triangulation component for fruit moving target detection, Ξ is the set of indexes for the triangulation component for fruit moving target detection, and λ k Let θ be the weighting coefficient of the triangular component for detecting the k-th moving fruit target. k ε is the median value of the triangular components for detecting the k-th moving fruit target. kThe sidelobe of the triangular component for detecting the k-th moving fruit target is u, which is the step function, and m and n are the x and y coordinates of the pixel, respectively.

[0080] Step 2: Establish the objective function for fruit moving target detection by analyzing the first few frames of the video. Introduce Lagrange daily numbers to ensure that the weight coefficients of the triangular components for fruit moving target detection meet the normalization condition. The specific determination method is as follows:

[0081] Suppose we use the first T frames of the video to detect moving fruit targets to establish a Lagrange daily number γ to realize λ. k The objective function min{F} for fruit moving target detection that satisfies the normalization condition is calculated using the following formula:

[0082]

[0083] Among them, setting Θ is the time set constructed from the previous T frames;

[0084] Step 3: After differentiating the objective function for fruit moving target detection with respect to the weight coefficients of the triangular components of fruit moving target detection to zero, obtain the weight coefficients of the triangular components of fruit moving target detection that minimize the objective function for fruit moving target detection. The specific method for determining this is as follows:

[0085] By the F to the λ k Differentiation yields the following specific calculation formula:

[0086]

[0087] By the F to the λ k There exists an extremum such that min{F} exists, which shows that... Therefore, we can obtain λ that satisfies min{F}. k The specific calculation formula is as follows:

[0088]

[0089] Step 4: Differentiate the objective function for fruit motion target detection with respect to the median value of the triangular components of fruit motion target detection until it becomes zero. This yields the median value of the triangular components that minimizes the objective function. The specific method for determining this is as follows:

[0090] By the F to the θ k Differentiation yields The specific calculation formula is as follows:

[0091]

[0092] Among them, to simplify the representation, let

[0093] By the F to the θ k There exists an extremum such that min{F} exists, which shows that... If the values ​​of H2 and H3 are small and When, then we can obtain θ that satisfies min{F} under this condition. k The specific calculation formula is as follows:

[0094]

[0095] Where max is the function that takes the maximum value, and abs is the function that takes the positive value.

[0096] Step 5: After differentiating the objective function for fruit motion target detection with respect to the sidelobes of the triangular components of fruit motion target detection to zero, obtain the sidelobes of the triangular components of fruit motion target detection that minimize the objective function for fruit motion target detection. The specific method for determining this is as follows:

[0097] By the F to the ε k The opposite number σ k Differentiation yields The specific calculation formula is as follows:

[0098]

[0099] For simplification, we can set...

[0100] By the F to the σ k and ε k There exists an extremum such that min{F} exists, which shows that... and Therefore, we can conclude that:

[0101]

[0102]

[0103] in,

[0104] Step 6: By introducing specific calculation formulas for the weight coefficients, median values, and sidelobes of the triangular components for fruit motion target detection that satisfy the objective function of fruit motion target detection, iterate repeatedly until the sum of the differences between the weight coefficients of the triangular components in the current round and the previous round is less than or equal to the weight coefficient difference threshold of the triangular components. Then, output the recommended values ​​for the weight coefficients, median values, and sidelobes of the triangular components for fruit motion target detection. The specific determination method is as follows: Figure 2 The steps shown are as follows:

[0105] Step 6-1: Set the iteration count h to 0, and statistically analyze the video of fruit movement targets. Figure 3 The image shows the grayscale values ​​of different pixels in the first T=3 frames, where the image size is 240×240 pixels. The average value of similar grayscale values, from highest to lowest frequency, is set as θ. k (h=0), and set θ k The number of occurrences of similar gray values ​​divided by num(Θ)·num(Ω) is set as λ. k (h=0), where num is a function to calculate the number of elements in the set, by using θ k (h=0), by satisfying the existence of ε in min{F} k The specific calculation formula for ε is given. k (h=0);

[0106] Step 6-2: Replace h with h+1, and change λ k (h-1) and ε k Substituting (h-1) into θ that satisfies min{F} k The specific calculation formula for θ is given. k (h);

[0107] Step 6-3: Place λ k (h-1) and θ k (h) Substitute ε that satisfies min{F} k The specific calculation formula for ε is given. k (h);

[0108] Step 6-4: ε k (h) and θ k (h) Substitute λ that satisfies min{F} k The specific calculation formula for λ is given. k (h);

[0109] Step 6-5: Place λ k (h) and λ k Compare (h-1), if If the condition is met, proceed to step 6-6; otherwise, proceed to step 6-2. The threshold δ for the difference in the weight coefficients of the triangular components can be set to 0.04.

[0110] Step 6-6: Place λ k (h), θ k (h) and ε k (h) are recommended weighting coefficients for the triangular components of fruit moving target detection. Median Recommended Value and sidelobe recommendation value Output;

[0111] Figure 4 The text presents the differences in the weight coefficients of the triangular components when using five triangular components for matching in different iterations. This can be seen from... Figure 4 It can be seen that when the number of iterations reaches 15, the difference in the weight coefficients of the triangular components gradually stabilizes, and in the detection of moving targets in fruit, the Gaussian components are also referenced, and generally 5 components are used as the matching reference.

[0112] Step 7: In the visual appearance detection, the recommended values ​​of the weight coefficients, median values, and sidelobe values ​​of the triangular components corresponding to each pixel in the fruit moving target video are used for matching detection in the triangular component probability density function. The specific determination method is as follows:

[0113] When detecting moving targets in fruit for visual appearance inspection, for a pixel (i,j) within frame l of a video, it is recommended to first use the triangulation component k′ used for that pixel in frame l-1. To perform a matching detection, the judgment is made through the probability density function of the triangular components. Limit the value to Within the region; considering the discrepancy between the median recommendation and the sidelobe recommendation, but because If the value is not in the above-mentioned region, it takes a negative value; therefore, in practical applications, the model can be corrected by detecting negative values. If the match is found, it is considered a successful match; otherwise, a recommendation value is made based on different weighting coefficients. Pass through in descending order Perform a matching check, where k∈Ξ, if there exists some k such that... If true, then pixel (i,j) uses the k-th fruit moving target detection triangulation component; if no k exists that can make true... If true, the gray value of the pixel (i,j) is set as the new triangular component, its weight coefficient is set, and the weight coefficients of other triangular components are adjusted.

[0114] Any parts not explicitly stated in this specific embodiment can be implemented using existing technologies.

[0115] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for detecting moving targets in fruit for visual appearance inspection, characterized in that, Includes the following steps: Step 1: Construct the probability density function of the triangular component corresponding to a certain pixel in the video for detecting moving fruit targets in visual appearance detection, using the weight coefficients of the triangular component for detecting moving fruit targets, the median value of the triangular component for detecting moving fruit targets, the side lobes of the triangular component for detecting moving fruit targets, and the gray value of a certain pixel at a certain time as independent variables. Step 2: Establish the objective function for detecting moving fruit targets by detecting the first few frames of the video, and introduce the Lagrange daily number to ensure that the weight coefficients of the triangular components for detecting moving fruit targets meet the normalization condition. Step 3: After differentiating the objective function for fruit moving target detection with respect to the weight coefficients of the triangular components of fruit moving target detection to zero, obtain the weight coefficients of the triangular components of fruit moving target detection that make the objective function for fruit moving target detection have a minimum value; Step 4: Obtain the intermediate value of the triangular components of the fruit moving target detection when the derivative of the objective function of fruit moving target detection with respect to the intermediate value of the triangular components of fruit moving target detection is zero; Step 5: Obtain the sidelobe of the fruit moving target detection triangular component when the objective function for fruit moving target detection has a minimum value by taking the derivative of the objective function with respect to the sidelobe of the fruit moving target detection triangular component as zero. Step 6: By introducing specific calculation formulas for the weight coefficients, median values, and side lobes of the fruit motion target detection triangular components that satisfy the objective function of fruit motion target detection, iterate repeatedly until the sum of the differences between the weight coefficients of the current round and the previous round of fruit motion target detection triangular components is less than or equal to the triangular component weight coefficient difference threshold, thereby outputting the recommended values ​​for the weight coefficients, median values, and side lobes of the fruit motion target detection triangular components; Step 7: In visual appearance detection, the recommended values ​​of the weight coefficients, median values, and sidelobe values ​​of the triangular components corresponding to each pixel in the fruit moving target video are used for matching detection in the triangular component probability density function.

2. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 1 specifically includes: The video is set to detect moving targets of fruit in visual appearance inspection at the pixel level. And the time is grayscale value The corresponding triangular component probability density function is The specific calculation formula is as follows: ； in, The triangular components for detecting moving targets of fruit are numbered. This is the set of numbers for the triangular components used in detecting moving targets on fruit. For the first The weighting coefficients of the triangular components for detecting moving targets in individual fruits. For the first The median value of the triangular components for detecting moving targets of individual fruits. For the first Side lobes of the triangular component for detecting moving targets in individual fruits. and These are the x and y coordinates of the pixel, respectively.

3. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 2 specifically includes: In a video used to detect the movement of fruit, the front... Frames are used to establish the introduction of Lagrange daily numbers To achieve Objective function for detecting moving fruit targets that meets normalization conditions The specific calculation formula is as follows: ； Among them, setting , For the front A time set for frame construction.

4. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 3 specifically includes: Depend on right Differentiation yields the following specific calculation formula: ； From the above Regarding the There exist extrema that satisfy Existence, knowability Therefore, we can conclude that: ； Considering After sorting, we get: ； Therefore, satisfaction can be obtained. Existing The specific calculation formula is as follows: 。 5. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 4 specifically includes: Depend on right Differentiation yields The specific calculation formula is as follows: Among them, to simplify the representation, let ; Can Approximately: From the above Regarding the There exist extrema that satisfy Existence, knowability Therefore, we can conclude that: make , , , , , as well as Therefore, we can conclude that: After sorting, we can obtain: make , , , , as well as Thus, the aforementioned The specific calculation formula for a quintic unary equation with variable A is as follows: set up , , , ,like and The value is small and Then, under these conditions, it can be obtained that... Existing The specific calculation formula is as follows: in, .

6. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 5 specifically includes: Depend on right the opposite number Differentiation yields The specific calculation formula is as follows: ； For simplification, we can set... ; After simplification, we can approximate the following: ； From the above Regarding the and There exist extrema that satisfy Existence, knowability and Therefore, we can conclude that: ； make , as well as ,Depend on and All are greater than 0, and It can be known that , and All are greater than 0, therefore we can obtain The specific calculation formula for a quartic equation with one variable is as follows: ； Depend on Satisfaction can be obtained Existing The specific calculation formula is as follows: 。 7. The method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, The specific steps of step 6 are as follows: Step 6-1: Set the number of iterations The value is 0, indicating that in the video of the fruit movement target, the previous value is 0. The grayscale values ​​of different pixels in the frame are averaged by taking the most frequent grayscale values ​​from the highest to the lowest, and then set as the average value. and the Divide the number of times similar gray values ​​appear by Set as ,in A function to calculate the number of elements in a set, by... By satisfying Existing The specific calculation formula will be given. ; Step 6-2: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require replace ,Will and Substitution satisfies Existing The specific calculation formula is given. ; Step 6-3: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] and Substitution satisfies Existing The specific calculation formula is given. ; Step 6-4: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] and Substitution satisfies Existing The specific calculation formula is given. ; Step 6-5: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require and If a comparison is made, Then proceed to step 6-6; otherwise, proceed to step 6-2. This is the threshold for the difference in weighting coefficients of the triangular components; Step 6-6: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , and Recommended values ​​for the weighting coefficients of the triangular components for fruit moving target detection. Recommended median value and sidelobe recommendation value Output the results.

8. A method for detecting moving targets in fruit based on visual appearance inspection according to claim 1, characterized in that, Step 7 specifically includes: In fruit moving target detection for visual appearance inspection, for video... Intra-frame pixels First, I recommend using The triangular component used by this pixel at frame time pass Perform a judgment to match and detect, if If the match is found, it is considered a successful match; otherwise, a recommendation value is made based on different weighting coefficients. Perform matching checks from largest to smallest; if a match exists... One of the middle Can make If true, then the pixel Adopting the first Trigonometric component analysis for detecting moving targets in individual fruits.

Images