Flotation condition recognition method and system based on weighted k-nearest neighbor algorithm
The flotation condition identification method based on the weighted K-nearest neighbor algorithm solves the problem of balancing efficiency and accuracy in image feature selection during foam flotation, achieving efficient and accurate flotation condition identification and optimizing the flotation process.
Patent Information
- Application Number
- CN202211344345.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-31
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2042-10-31
AI Technical Summary
In existing foam flotation technology, image feature selection methods are difficult to balance classification efficiency and accuracy, resulting in low flotation precision, substandard product quality, waste of reagents, and serious loss of resources.
A flotation condition identification method based on the weighted K-nearest neighbor algorithm is adopted. By setting fuzzy multi-neighbor particle settings, calculating entropy concepts and analyzing multiple relationships, a target evaluation function is constructed, and the feature subset with the highest classification accuracy is selected for flotation condition identification.
It improves the classification efficiency and accuracy of flotation condition identification, makes full use of the useful information provided by the features of the flotation image, and optimizes the flotation process.
Smart Images
Figure CN115578600B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of froth flotation, and discloses a flotation condition recognition method and system based on a weighted K nearest neighbor algorithm. BACKGROUND
[0002] Mineral resources are the material basis for the survival and development of human society, and are also an important guarantee for national security and economic development. Froth flotation is the most widely used mineral processing method, and almost all minerals can be separated by froth flotation. Due to the long industrial process of flotation, the complex mechanism, the large number of variables involved in the process, and the strong coupling, the flotation process has always relied on manual observation of the surface froth state of the flotation tank to adjust the dosage of reagents. This subjective method will affect the accuracy of the flotation condition recognition, and it is difficult to quantify and control the reagent dosage, to objectively evaluate and recognize the flotation froth state, and to achieve the overall optimization of the flotation process, resulting in low flotation precision, unqualified product quality, waste of flotation reagents, serious loss of mineral raw materials, and low resource recovery rate. Therefore, it is of great significance to study the surface visual features and realize the objective recognition of the froth state and the overall optimization of the flotation process.
[0003] In recent years, researchers have conducted a lot of research on froth image processing and froth surface visual feature extraction, and more and more extracted features are used for flotation condition recognition. However, these features are relatively single and low-dimensional, and only a small amount of information is available, or too many froth image features are used, resulting in information redundancy, which is not conducive to effective control of the flotation process. The previous froth image feature selection method only considers the correlation and redundancy between features, without fully considering the multiple relationships between froth image features, so that the useful information provided by the features cannot be fully utilized, thereby increasing the training difficulty and computational complexity of the classifier, and even reducing the condition recognition accuracy.
[0004] Therefore, the existing froth image feature selection method cannot balance the classification efficiency and classification accuracy, which is a technical problem to be solved at present. SUMMARY
[0005] The present application provides a flotation condition recognition method and system based on a weighted K nearest neighbor algorithm, which aims to solve the technical problem that the existing froth image feature selection method cannot balance the classification efficiency and classification accuracy.
[0006] One aspect of the present application relates to a flotation condition recognition method based on a weighted K nearest neighbor algorithm, comprising the following steps:
[0007] Fuzzy multi-neighbor particles are set for fuzzy and uncertain data in the froth image data set and different distributions of various features;
[0008] The concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, and the multiple relationships among the features of the froth image are fully considered under the fuzzy multi-granularity information theory framework, including correlation, redundancy, complementarity and interaction;
[0009] A target evaluation function is constructed according to the multiple relationships among the features, the function value of each feature is calculated, and the features are sorted;
[0010] The K-nearest neighbor algorithm is weighted by using the calculated feature function value;
[0011] The feature subset with the highest classification accuracy is selected as the optimal feature subset for the flotation condition recognition.
[0012] Further, the step of setting the fuzzy multi-neighborhood granule for the fuzzy and uncertain data in the froth image data set and the different distributions of various features comprises:
[0013] The froth video captured by the camera is equally spaced and cut, and one frame of the cut image constitutes an original sample set;
[0014] The relevant features are extracted from each froth image sample in the sample set to constitute a feature set;
[0015] The feature set is normalized and pretreated to eliminate the dimensional influence among different features;
[0016] The fuzzy multi-neighborhood granule is set for various features after eliminating the dimensional influence.
[0017] Further, the concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, and the multiple relationships among the features of the froth image are fully considered under the fuzzy multi-granularity information theory framework.
[0018] The concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, and the multiple relationships among the features of the froth image are fully considered under the fuzzy multi-granularity information theory framework.
[0019] Under the fuzzy multi-granularity information theory framework, the multiple relationships among the features of the froth image are fully considered, a feature target evaluation function is constructed to measure the contribution degree of the froth image features to the image classification, and the features are sorted, the feature with the largest correlation with the class D is first selected into the selected feature subset S, then the features with the largest feature target evaluation function are selected into the selected feature subset S in turn, until the features in the remaining candidate subset V are selected.
[0020] Further, the calculation formula of the characteristic target evaluation function is as follows:
[0021]
[0022] FMI(f i ; D) is the relevance of the current candidate characteristic f i to the category D, FMCI(f i ; f s ) is the complementarity of the current candidate characteristic f i to the selected characteristic subset S, FMCI(f i ; D|f s ) is the redundancy of the current candidate characteristic f i to the selected characteristic f s , FMCI(f j ; D|f i ) is the interaction of the remaining candidate characteristic f j to the current candidate characteristic f i , F is the characteristic set, S is the selected characteristic subset, and V is the remaining candidate characteristic subset.
[0023] Further, in the step of weighting the K-Nearest Neighbor algorithm by using the calculated characteristic function value, the weight of the current candidate characteristic f i is defined as the standard of the value of the target evaluation function of the current candidate characteristic f i , and the Euclidean distance in the K-Nearest Neighbor algorithm is weighted after the weight of each ranked characteristic is obtained.
[0024] The weight of the current candidate characteristic f i is as follows:
[0025]
[0026] wherein w i is the weight of the current candidate characteristic f i , is the value of the target evaluation function of the current candidate characteristic f i , and m is the total number of characteristics.
[0027] The weighting formula is as follows:
[0028]
[0029] wherein d(x, y) is the weighted Euclidean distance in the K-Nearest Neighbor algorithm, m is the total number of characteristics, and are the values of the sample x and the sample y with respect to the characteristic f i .
[0030] Another aspect of the present application relates to a flotation condition recognition system based on a weighted K nearest neighbor algorithm, comprising:
[0031] The setting module is configured to set fuzzy multi-neighborhood particles for fuzzy and uncertain data in the foam image dataset and different distributions of various features.
[0032] The first calculation module is configured to introduce the concept of entropy to calculate an amount of information provided by the fuzzy multi-neighborhood particles induced by the feature subset, and fully consider multiple relationships among the foam image features in the fuzzy multi-granularity information theory framework, the multiple relationships including correlation, redundancy, complementarity and interaction.
[0033] The second calculation module is configured to construct a target evaluation function according to the multiple relationships among the features, calculate a function value of each feature, and sort the features.
[0034] The weighting module is configured to weight the K nearest neighbor algorithm by using the calculated feature function values.
[0035] The selection module is configured to select a feature subset with the highest classification accuracy as an optimal feature subset for the flotation condition recognition.
[0036] Further, the setting module comprises:
[0037] The intercepting unit is configured to intercept the flotation foam video captured by the camera at equal intervals, and form an original sample set by using the intercepted image.
[0038] The extracting unit is configured to extract relevant features from each foam image sample in the sample set to form a feature set.
[0039] The preprocessing unit is configured to perform normalization preprocessing on the feature set to eliminate the influence of dimensions among different features.
[0040] The setting unit is configured to set fuzzy multi-neighborhood particles for various features after eliminating the influence of dimensions.
[0041] Further, the first calculation module comprises:
[0042] The first calculation unit is configured to introduce the concept of entropy to calculate an amount of information provided by the fuzzy multi-neighborhood particles induced by the feature subset, the amount of information including fuzzy multi-neighborhood mutual information of the sample set with respect to a feature subset A and a feature subset B, and fuzzy conditional mutual information of the sample set with respect to the feature subset A, a feature subset B and a feature subset C.
[0043] The construction unit is used for constructing a feature target evaluation function under the fuzzy multi-granularity information theory framework to measure the contribution of the foam image features to the image classification and performing feature sorting. The feature first entering the selected feature subset S is the feature having the largest correlation with the category D, and then the features having the largest feature target evaluation function are selected into the selected feature subset S in sequence until the features in the remaining candidate subset V are selected.
[0044] Further, in the second calculation module, the calculation formula of the feature target evaluation function is as follows:
[0045]
[0046] FMI(f i ; D) is the correlation of the current candidate feature f i with the category D, FMI(f i ; f s ) is the redundancy of the current candidate feature f i to the selected feature subset S, FMCI(f i ; D|f s ) is the complementarity of the current candidate feature f i to the selected feature f s , FMCI(f j ; D|f i ) is the interaction of the remaining unselected feature f j and the current candidate feature f i , F is the feature set, S is the selected feature subset, and V is the remaining candidate feature subset.
[0047] Further, in the weighting module, the weight of the current candidate feature f i is defined by taking the value of the feature target evaluation function of the current candidate feature f i as a standard, and the Euclidean distance in the K nearest neighbor algorithm is weighted after the weight of each sorted feature is obtained.
[0048] The weight of the current candidate feature f i is as follows:
[0049]
[0050] wherein w i is the weight of the current candidate feature f i , is the value of the feature target evaluation function of the current candidate feature f i , and m is the total number of features.
[0051] The weighting formula is as follows:
[0052]
[0053] wherein d(x, y) is a weighted Euclidean distance in K nearest neighbor algorithm, m is a total number of features, and are values of sample x and sample y about feature f i .
[0054] The present application has the following beneficial effects:
[0055] The present application provides a flotation condition recognition method and system based on weighted K nearest neighbor algorithm, which sets fuzzy multi-neighbor particles for fuzzy and uncertain data in bubble image data set and different distribution of various features; introduces the concept of entropy to calculate the information amount provided by fuzzy multi-neighbor particles induced by feature subsets, and fully considers the multiple relationships among bubble image features in the framework of fuzzy multi-granularity information theory, including correlation, redundancy, complementarity and interaction; constructs a target evaluation function according to the multiple relationships among features, calculates the function value of each feature, and sorts the features; weights K nearest neighbor algorithm by using the calculated feature function value; and selects the feature subset with the highest classification accuracy as the optimal feature subset for flotation condition recognition. The flotation condition recognition method and system based on weighted K nearest neighbor algorithm provided by the present application fully considers the multiple relationships among bubble image features, can fully utilize useful information provided by features, and has high classification efficiency and classification accuracy. BRIEF DESCRIPTION OF DRAWINGS
[0056] Figure 1 Fig. 1 is a flowchart of an embodiment of the flotation condition recognition method based on weighted K nearest neighbor algorithm provided by the present application;
[0057] Figure 2 Fig. 2 is a schematic diagram of condition samples in the flotation condition recognition method based on weighted K nearest neighbor algorithm provided by the present application, wherein (a)-(c), (d)-(f) and (g)-(i) are schematic diagrams of over-flotation, normal flotation and under-flotation condition samples, respectively;
[0058] Figure 3 Fig. 3 is a schematic diagram of multiple relationships among bubble image features in the flotation condition recognition method based on weighted K nearest neighbor algorithm provided by the present application;
[0059] Figure 4 Fig. 4 is a comparison diagram of classification accuracy of the flotation condition recognition method based on weighted K nearest neighbor algorithm provided by the present application and other four benchmark classifiers;
[0060] Figure 5 Fig. 5 is a comparison diagram of classification accuracy of the flotation condition recognition method based on weighted K nearest neighbor algorithm provided by the present application and other four benchmark classifiers; Figure 2A detailed flow chart of an embodiment of the step of setting fuzzy multi-neighborhood particles for fuzzy and uncertain data in the foam image data set and different distributions of various features shown in FIG. 1;
[0061] Figure 6 A detailed flow chart of an embodiment of the step of constructing a target evaluation function according to multiple relationships between features, calculating a function value of each feature, and ranking the features shown in FIG. 2; Figure 2 A detailed flow chart of an embodiment of the step of constructing a target evaluation function according to multiple relationships between features, calculating a function value of each feature, and ranking the features shown in FIG. 2;
[0062] Figure 7 A functional block diagram of an embodiment of the flotation condition recognition system based on the weighted K-nearest neighbor algorithm provided by the present application;
[0063] Figure 8 A detailed flow chart of an embodiment of the step of constructing a target evaluation function according to multiple relationships between features, calculating a function value of each feature, and ranking the features shown in FIG. 2; Figure 7 A functional block diagram of an embodiment of the setting module shown in FIG. 3;
[0064] Figure 9 A functional block diagram of an embodiment of the first calculation module shown in FIG. 4; Figure 7 A functional block diagram of an embodiment of the first calculation module shown in FIG. 4;
[0065] BRIEF DESCRIPTION OF THE DRAWINGS
[0066] 10, setting module; 20, first calculation module; 30, second calculation module; 40, weighting module; 50, selection module; 11, intercepting unit; 12, extracting unit; 13, preprocessing unit; 14, setting unit; 21, first calculation unit; 22, construction unit. DETAILED DESCRIPTION
[0067] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings in the specification and specific embodiments.
[0068] As shown in FIG. 1, the first embodiment of the present application proposes a flotation condition recognition method based on the weighted K-nearest neighbor algorithm, which includes the following steps: Figure 4 Step S100, setting fuzzy multi-neighborhood particles for fuzzy and uncertain data in the foam image data set and different distributions of various features.
[0069] A flotation foam video is obtained by a camera, and the obtained flotation foam video is intercepted at equal intervals to form an original sample set. Fuzzy multi-neighborhood particles are set for fuzzy and uncertain data in the foam image data set and different features.
[0070] A flotation foam video is obtained by a camera, and the obtained flotation foam video is intercepted at equal intervals to form an original sample set. Fuzzy multi-neighborhood particles are set for fuzzy and uncertain data in the foam image data set and different features.
[0071] A flotation foam video is obtained by a camera, and the obtained flotation foam video is intercepted at equal intervals to form an original sample set. Fuzzy multi-neighborhood particles are set for fuzzy and uncertain data in the foam image data set and different features.Step S200, the concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, and the multiple relationships between the features of the foam image are fully considered under the framework of fuzzy multi-granularity information theory.
[0072] The concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, which is the fuzzy multi-neighborhood mutual information FMI(A; B) of the sample set U with respect to the feature subset A and the feature subset B, and the fuzzy conditional mutual information FMCI(B; C|A) of the feature subset B and the feature subset C given the feature subset A. The multiple relationships between the features of the foam image are fully considered under the framework of fuzzy multi-granularity information theory, wherein, as shown in the formula (1), the multiple relationships include the correlation Rel, the redundancy Red, the complementarity Com and the interaction Int. Figure 3
[0073] Step S300, a target evaluation function is constructed according to the multiple relationships between the features, the function value of each feature is calculated, and the features are sorted.
[0074] According to the multiple relationships between the features of the foam image proposed under the framework of fuzzy multi-granularity information theory, a feature target evaluation function is constructed to measure the contribution degree of the features of the foam image to the image classification, the function value of each feature is calculated, and the features are sorted.
[0075] The first feature to enter the selected feature subset S is the feature with the largest correlation with the category D, and then the features with the largest correlation with the category D are selected into the selected feature subset S in turn, until the features in the remaining candidate subset V are selected, and the calculation formula of the feature target evaluation function is as follows:
[0076]
[0077] In the formula (1), FMI(f i ; D) is the correlation of the current candidate feature f i with the category D, FMI(f i ; f s ) is the redundancy of the current candidate feature f i to the selected feature subset S, FMCI(f i ; D|f s ) is the complementarity of the current candidate feature f i to the selected feature f s , and FMCI(f j ; D|f i ) is the interaction of the remaining unselected feature f j with the current candidate feature f i the interaction of the features, F is the feature set, S is the selected feature subset, and V is the remaining candidate feature subset.
[0078] Step S400: Weighting the K-Nearest Neighbor algorithm using the calculated feature function value.
[0079] The weight of the current candidate feature f i is defined as the value of the target evaluation function. i After obtaining the weight of each ranked feature, the Euclidean distance in the K-Nearest Neighbor algorithm is weighted.
[0080] The weight of the current candidate feature f i is:
[0081]
[0082] In formula (2), w i is the weight of the current candidate feature f i , is the value of the target evaluation function of the current candidate feature f i , and m is the total number of features.
[0083] The weighting formula is:
[0084]
[0085] In formula (3), d(x, y) is the weighted Euclidean distance in the K-Nearest Neighbor algorithm, m is the total number of features, and are the values of the sample x and the sample y with respect to the feature f i .
[0086] Step S500: Selecting the feature subset with the highest classification accuracy as the optimal feature subset for flotation condition recognition.
[0087] After adding one feature from the remaining candidate feature subset V into the selected feature subset S, the weighted K-Nearest Neighbor algorithm is used to classify the foam image, and the classification accuracy of the selected feature subset S is recorded each time. Finally, the feature subset with the highest classification accuracy is selected as the optimal feature subset for foam image condition recognition, and is compared with the other four benchmark classifiers. The effectiveness of the method proposed in this embodiment is determined by comparing the qualitative and quantitative evaluations of the five classifiers on the same data set, such as Figure 4The other four benchmark classifiers are SVM (Support Vector Machine), CART (Classification and Regression Tree), NB (Naive Bayes), and KNN (k-NearestNeighbor), respectively. The weighted KNN (weighted k-NearestNeighbor) is the method proposed in the embodiment.
[0088] Compared with the prior art, the flotation condition recognition method based on the weighted K nearest neighbor algorithm provided in the embodiment sets fuzzy multi-neighbor particles for fuzzy and uncertain data in the froth image data set and different distributions of various features; introduces the concept of entropy to calculate the amount of information provided by the fuzzy multi-neighbor particles induced by the feature subset, and fully considers the multiple relationships between the froth image features in the fuzzy multi-granularity information theory framework, including correlation, redundancy, complementarity, and interaction; constructs a target evaluation function according to the multiple relationships between the features, calculates the function value of each feature, and sorts the features; weights the K nearest neighbor algorithm using the calculated feature function value; and selects the feature subset with the highest classification accuracy as the optimal feature subset for flotation condition recognition. The flotation condition recognition method based on the weighted K nearest neighbor algorithm provided in the embodiment fully considers the multiple relationships between the froth image features, can fully utilize the useful information provided by the features, and has high classification efficiency and classification accuracy.
[0089] Further, see Figure 5 , Figure 5 for Figure 2 an embodiment of the step S100 shown in the flowchart, in the embodiment, the step S100 includes:
[0090] The step S110, the flotation froth video photographed by the camera is intercepted at equal intervals, and a frame of image intercepted constitutes an original sample set.
[0091] The flotation froth video photographed by the camera is collected, the flotation froth video photographed by the camera is intercepted at equal intervals, and a frame of image intercepted constitutes an original sample set U={x1, x2, …, x n}, the sample set U={x1, x2, …, x n} contains 324 froth images with three different flotation conditions of over-flotation, normal flotation, and under-flotation, and three sample pictures are selected as shown in Figure 2 , and each image has a size of 692 pixel values x 518 pixel values.
[0092] Step S120, extracting relevant features from each foam image sample in the sample set to form a feature set.
[0093] Extracting relevant features from each foam image sample in the sample set U = {x1, x2, …, x n} to form a feature set F = {f1, f2, …, f m}, including 11 static features, 3 dynamic features and 6 texture features.
[0094] Step S130, normalizing the feature set to eliminate the influence of the dimensions of different features.
[0095] Normalizing the data in the obtained feature set F = {f1, f2, …, f m} to eliminate the influence of the dimensions of different features.
[0096] Step S140, setting fuzzy multi-neighborhood granules for each type of feature after eliminating the influence of the dimensions.
[0097] Using a multi-neighborhood radius set instead of a uniform neighborhood value, and then setting fuzzy multi-neighborhood granules for each type of feature by adaptively adjusting the fuzzy similarity relation.
[0098] Compared with the prior art, the flotation condition recognition method based on the weighted K nearest neighbor algorithm provided in the embodiment, by equally spacedly intercepting the flotation foam video shot by the camera, forming an original sample set from the intercepted image, normalizing the feature set to eliminate the influence of the dimensions of different features, constructing a target evaluation function according to the multiple relationships between the features, calculating the function value of each feature and sorting the features, weighting the K nearest neighbor algorithm by using the calculated feature function value, and setting fuzzy multi-neighborhood granules for each type of feature after eliminating the influence of the dimensions, the method fully considers the multiple relationships between the foam image features, can fully utilize the useful information provided by the features, and has high classification efficiency and classification accuracy.
[0099] Preferably, referring to Figure 6 , Figure 6 is Figure 2 the detailed flowchart of an embodiment of step S200 shown in the figure, in the embodiment, step S200 includes:
[0100] Step S210, introducing the concept of entropy to calculate the amount of information provided by the fuzzy multi-neighborhood granules induced by the feature subset.
[0101] The concept of entropy is introduced to calculate the information amount provided by the fuzzy multi-neighborhood granule induced by the feature subset, such as fuzzy multi-neighborhood entropy FME, fuzzy multi-neighborhood joint entropy FMJE, and fuzzy multi-neighborhood conditional entropy FMC. Based on this, mutual information is introduced to measure the correlation between features and classes and between features, which are fuzzy multi-neighborhood mutual information FMI(A; B) of the sample set U with respect to the feature subset A and the feature subset B, and the fuzzy conditional mutual information FMCI(B; C|A) of the feature subset B, the feature subset C and the feature subset A in the sample set U.
[0102] Step S220, under the framework of fuzzy multi-granularity information theory, fully considering the multiple relationships between the features of the foam image, a feature target evaluation function is constructed to measure the contribution of the foam image features to image classification, and the features are sorted. The first feature to enter the selected feature subset S is the feature with the largest correlation with the class D. Then the feature with the largest feature target evaluation function is selected into the selected feature subset S, and the process is repeated until all the features in the remaining candidate subset V are selected.
[0103] According to the multiple relationships between the features of the foam image under the framework of fuzzy multi-granularity information theory, that is, the correlation Rel, the redundancy Red, the complementarity Com, and the interaction Int, a feature target evaluation function is constructed to measure the contribution of the foam image features to image classification, and the multiple relationship diagram is shown in Figure 3 According to the feature target evaluation function , the features are sorted to obtain a sorted feature set. According to the sorting, a heuristic search algorithm is used to generate multiple feature subsets. The heuristic search algorithm is a sequential forward selection algorithm, which specifically starts from an empty set, adds the feature ranked first to the empty set to form the first feature subset, and then selects the feature ranked i from the sorted feature set to add to the i-1th feature subset to form the i th feature subset until all the features in the sorted feature set are selected.
[0104] Compared with the prior art, the method for identifying the flotation condition based on the weighted K nearest neighbor algorithm provided in the embodiment calculates the information amount provided by the fuzzy multi-neighbor particles induced by the feature subset by introducing the concept of entropy, fully considers the multiple relationships among the features of the froth image in the fuzzy multi-granularity information theory framework, constructs a feature target evaluation function to measure the contribution degree of the features of the froth image to the image classification, and performs feature sorting, and the feature that first enters the selected feature subset S is the feature with the largest correlation with the class D, and then the features with the largest feature target evaluation function are selected into the selected feature subset S in turn until the features in the remaining candidate subset V are selected.
[0105] As shown in Figure 7 , the feature target evaluation function is constructed according to the multiple relationships among the features, the function value of each feature is calculated, and the features are sorted. Figure 7 The function block diagram of an embodiment of the flotation condition identification system based on the weighted K nearest neighbor algorithm provided in the application, in the embodiment, the flotation condition identification system based on the weighted K nearest neighbor algorithm comprises a setting module 10, a first calculation module 20, a second calculation module 30, a weighting module 40 and a selection module 50, wherein the setting module 10 is used for setting the fuzzy multi-neighbor particles of the fuzzy and uncertain data in the froth image data set and the different distribution of various features; the first calculation module 20 is used for calculating the information amount provided by the fuzzy multi-neighbor particles induced by the feature subset by introducing the concept of entropy, and fully considering the multiple relationships among the features of the froth image in the fuzzy multi-granularity information theory framework, the multiple relationships including correlation, redundancy, complementarity and interaction; the second calculation module 30 is used for constructing a target evaluation function according to the multiple relationships among the features, calculating the function value of each feature, and sorting the features; the weighting module 40 is used for weighting the K nearest neighbor algorithm by using the calculated feature function value; and the selection module 50 is used for selecting the feature subset with the highest classification accuracy as the optimal feature subset for identifying the flotation condition.
[0106] The setting module 10 obtains the flotation froth video shot by the camera, and equally spacedly intercepts the shot flotation froth video to form an original sample set by one frame of image.
[0107] The first calculation module 20 introduces the concept of entropy to calculate the information amount provided by the fuzzy multi-neighborhood granules induced by the feature subsets, which are respectively fuzzy multi-neighborhood mutual information FMI(A;B) of the sample set U with respect to the feature subsets A and B, and fuzzy conditional mutual information FMCI(B;C|A) of the feature subsets B and C given the feature subset A. The multiple relationships among the features of the foam image are fully considered under the framework of fuzzy multi-granularity information theory, wherein the multiple relationships include correlation Rel, redundancy Red, complementarity Com and interaction Int.
[0108] The second calculation module 30 constructs a feature target evaluation function according to the multiple relationships among the features of the foam image proposed under the framework of fuzzy multi-granularity information theory to measure the contribution of the features of the foam image to image classification, calculates the function value of each feature, and performs feature sorting.
[0109] The feature with the largest correlation with the category D is firstly selected into the selected feature subset S, and then the features with the largest correlation with the category D are sequentially selected into the selected feature subset S until the features in the remaining candidate subset V are selected.
[0110]
[0111] In formula (4), FMI(f i ;D) is the correlation of the current candidate feature f i with the category D, FMI(f i ;f s ) is the redundancy of the current candidate feature f i to the selected feature subset S, FMCI(f i ;D|f s ) is the complementarity of the current candidate feature f i to the selected feature f s , FMCI(f j ;D|f i ) is the interaction of the remaining unselected feature f j with the current candidate feature f i , F is the feature set, S is the selected feature subset, and V is the remaining candidate feature subset.
[0112] The weighting module 40 defines the weight of the current candidate feature f i based on the value of the feature target evaluation function of the current candidate feature f i , and obtains the weight of each sorted feature, and then weights the Euclidean distance in the K-nearest neighbor algorithm.
[0113] The weight of the current candidate feature f i is:
[0114]
[0115] In formula (5), w i For the current candidate feature f i The weight, For the current candidate feature f i The value of the objective evaluation function, where m is the total number of features.
[0116] The weighted formula is:
[0117]
[0118] In formula (6), d(x, y) is the weighted Euclidean distance in the K-nearest neighbor algorithm, and m is the total number of features. and They are samples x and y with respect to feature f, respectively. i The value of .
[0119] In the selection module 50, for each feature added to the remaining candidate feature subset V into the selected feature subset S, the weighted K-nearest neighbor algorithm is used to classify the foam image. The classification accuracy of the selected feature subset S is recorded each time. Finally, the feature subset with the highest classification accuracy is selected as the optimal feature subset for foam image condition identification, and compared with four other benchmark classifiers. The effectiveness of the proposed method in this embodiment is determined by comparing the qualitative and quantitative evaluations of the five classifiers on the same dataset. Figure 4 As shown, the other four benchmark classifiers are SVM (Support Vector Machine), CART (Classification and Regression Tree), NB (NaiveBayes), and KNN (k-Nearest Neighbor). Weighted KNN (Weighted k-Nearest Neighbor) is the method proposed in the example.
[0120] Compared with the prior art, the flotation condition recognition system based on the weighted K nearest neighbor algorithm provided in the embodiment adopts the setting module 10, the first calculation module 20, the second calculation module 30, the weighting module 40 and the selection module 50, sets the fuzzy multi-neighborhood particles for the fuzzy and uncertain data in the foam image data set and the different distributions of various features, calculates the information amount provided by the fuzzy multi-neighborhood particles induced by the feature subsets by introducing the concept of entropy, fully considers the multiple relationships between the foam image features in the fuzzy multi-granularity information theory framework, the multiple relationships including correlation, redundancy, complementarity and interaction, constructs the target evaluation function according to the multiple relationships between the features, calculates the function value of each feature and sorts the features, weights the K nearest neighbor algorithm by using the calculated feature function value, and selects the feature subset with the highest classification accuracy as the optimal feature subset for the flotation condition recognition. The flotation condition recognition system based on the weighted K nearest neighbor algorithm provided in the embodiment fully considers the multiple relationships between the foam image features, can fully utilize the useful information provided by the features, and has high classification efficiency and classification accuracy.
[0121] Further, see Figure 8 , Figure 8 for Figure 7 the function module schematic diagram of the setting module embodiment shown in the figure, in the embodiment, the setting module 10 includes the intercepting unit 11, the extracting unit 12, the preprocessing unit 13 and the setting unit 14, wherein the intercepting unit 11 is used for intercepting the flotation foam video shot by the camera at equal intervals, and one frame of image intercepted constitutes an original sample set; the extracting unit 12 is used for extracting relevant features from each foam image sample in the sample set to constitute a feature set; the preprocessing unit 13 is used for normalizing the feature set for preprocessing, eliminating the influence of the dimensions between different features; and the setting unit 14 is used for setting the fuzzy multi-neighborhood particles for various features with the influence of the dimensions eliminated.
[0122] The intercepting unit 11 collects the flotation foam video shot by the camera, intercepts the flotation foam video shot by the camera at equal intervals, and one frame of image intercepted constitutes an original sample set U={x1, x2,…, x n}, the sample set U={x1, x2,…, x n} contains 324 foam images with three different flotation conditions of over-flotation, normal flotation and under-flotation, and each selects 3 sample pictures as shown in Figure 2 , the size of each image is 692 pixel values x 518 pixel values.
[0123] The extracting unit 12 extracts relevant features from each foam image sample of the sample set U={x1, x2,…, x n}, and a total of 20 constitutes a feature set F={f1, f2,…, fm}, wherein, including static characteristics 11, dynamic characteristics 3, texture characteristics 6.
[0124] The pre-processing unit 13 carries out normalization pre-processing on the data in the obtained feature set F = {f1, f2, …, f m}, eliminating the influence of the dimension between different features.
[0125] The setting unit 14 uses a multi-neighborhood radius set instead of a uniform neighborhood value, and then sets fuzzy multi-neighborhood granules for various features through adaptive adjustment of the fuzzy similarity relation.
[0126] Compared with the prior art, the setting module 10 adopts the intercepting unit 11, the extracting unit 12, the pre-processing unit 13 and the setting unit 14, the flotation froth video captured by the camera is intercepted at equal intervals, one frame of image after interception is used to form an original sample set, the feature set is subjected to normalization pre-processing, the influence of the dimension between different features is eliminated, a target evaluation function is constructed according to the multiple relationships between features, the function value of each feature is calculated, and the features are sorted, the K nearest neighbor algorithm is weighted by using the calculated feature function value, and fuzzy multi-neighborhood granules are set for various features after eliminating the influence of the dimension. The flotation froth condition recognition system based on the weighted K nearest neighbor algorithm provided by the present application fully considers the multiple relationships between the features of the froth image, can make full use of the useful information provided by the features, and has high classification efficiency and classification accuracy.
[0127] Preferably, referring to Figure 9 , Figure 9 is Figure 7 the functional module schematic diagram of the first calculation module embodiment shown in the first calculation module, in the present embodiment, the first calculation module 20 includes a first calculation unit 21 and a construction unit 22, wherein the first calculation unit 21 is used for introducing the concept of entropy to calculate the amount of information provided by the fuzzy multi-neighborhood granules induced by the feature subsets, the amount of information includes fuzzy multi-neighborhood mutual information of the sample set with respect to the feature subset A and the feature subset B, and fuzzy conditional mutual information of the feature subset B and the feature subset C of the sample set under the given feature subset A; the construction unit 22 is used for fully considering the multiple relationships between the features of the froth image under the fuzzy multi-granularity information theory framework, constructing a feature target evaluation function to measure the contribution degree of the features of the froth image to image classification, and performing feature sorting, the feature that first enters the selected feature subset S is the feature that has the largest correlation with the category D, then the features with the largest feature target evaluation function are selected into the selected feature subset S in sequence, until the features in the remaining candidate subset V are selected completely.
[0128] The first computing unit 21 introduces the concept of entropy to calculate the amount of information provided by the fuzzy multi-neighborhood particles induced by the feature subset, such as fuzzy multi-neighborhood entropy FME, fuzzy multi-neighborhood joint entropy FMJE, and fuzzy multi-neighborhood conditional entropy FMC. Based on this, mutual information is introduced to measure the correlation between features and classes and between features, which are fuzzy multi-neighborhood mutual information FMI(A; B) of the sample set U with respect to the feature subset A, the feature subset B, and the fuzzy conditional mutual information FMCI(B; C|A) of the feature subset B, the feature subset C, and the feature subset A under the condition that the sample set U is given.
[0129] The construction unit 22 constructs a feature target evaluation function according to the multiple relationships between the foam image features, i.e. correlation Rel, redundancy Red, complementarity Com, and interaction Int, under the framework of fuzzy multi-granularity information theory to measure the contribution of the foam image features to image classification, and the multiple relationship is shown in the schematic diagram Figure 3 , and the feature sorting is performed according to the feature target evaluation function to obtain a sorted feature set. According to the sorting, a plurality of feature subsets are generated using a heuristic search algorithm. The heuristic search algorithm is a sequential forward selection algorithm, and the forward selection algorithm specifically starts from an empty set, adds the feature ranked first to the empty set to form the first feature subset, and then selects the feature ranked i from the sorted feature set to add to the i-1th feature subset to form the i-th feature subset until all features in the sorted feature set are selected.
[0130] Compared with the prior art, the first computing module 20 adopts the first computing unit 21 and the construction unit 22, calculates the amount of information provided by the fuzzy multi-neighborhood particles induced by the feature subset by introducing the concept of entropy, fully considers the multiple relationships between the foam image features under the framework of fuzzy multi-granularity information theory, constructs a feature target evaluation function to measure the contribution of the foam image features to image classification, and performs feature sorting. The feature that is most relevant to the class D is first entered into the selected feature subset S, and then the feature with the largest feature target evaluation function is selected into the selected feature subset S in turn until all features in the remaining candidate subset V are selected. The flotation condition recognition system based on the weighted K nearest neighbor algorithm provided in the embodiment fully considers the multiple relationships between the foam image features, can fully utilize the useful information provided by the features, and has high classification efficiency and classification accuracy.
[0131] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the foregoing description without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims be interpreted as including all such variations and modifications as fall within the spirit and scope of the application. It is further intended that the disclosure of all such modifications and variations be included within the scope of the application, the terms used herein being defined solely for purposes of the description being applied thereto unless otherwise indicated.
Claims
1. A flotation operating condition recognition method based on a weighted K-nearest neighbor algorithm, characterized by, It comprises the following steps: The fuzzy multi-neighborhood particles are set for the fuzzy and uncertain data in the foam image dataset and different distributions of various features, including: The video of the flotation foam shot by the camera is equally spaced to extract a frame of image to form an original sample set; Features are extracted from each foam image sample in the sample set to form a feature set; The feature set is normalized to eliminate the dimension influence between different features; The fuzzy multi-neighborhood particles are set for various features after eliminating the dimension influence; The concept of entropy is introduced to calculate the information provided by the fuzzy multi-neighborhood particles induced by the feature subset, and the multiple relationships between the foam image features are fully considered under the fuzzy multi-granularity information theory framework, including the correlation, redundancy, complementarity and interaction, including: The concept of entropy is introduced to calculate the amount of information provided by the fuzzy multi-neighborhood granules induced by the feature subsets, which includes the fuzzy multi-neighborhood mutual information of the sample set with respect to the feature subset A, the feature subset B, and the fuzzy conditional mutual information of the sample set with respect to the feature subset B given the feature subset A ; A feature target evaluation function is constructed to measure the contribution of the foam image features to the image classification under the fuzzy multi-granularity information theory framework, and the features are sorted, the feature with the largest correlation with the class D is first selected into the selected feature subset S, and then the features with the largest feature target evaluation function are selected into the selected feature subset S, until the features in the remaining candidate subset V are selected, and the calculation formula of the feature target evaluation function is as follows: in, For current candidate features With category The correlation, For current candidate features For the selected feature subset Redundancy, For current candidate features For selected features Complementarity Remaining unselected features With current candidate features Interactivity For feature set, For the selected feature subset, This is a subset of the remaining candidate features; The target evaluation function is constructed according to the multiple relationships between the features, the function value of each feature is calculated, and the features are sorted; The K-nearest neighbor algorithm is weighted by using the calculated feature function value; The feature subset with the highest classification accuracy is selected as the optimal feature subset for flotation condition recognition.
2. The method of claim 1, wherein the weighted K-nearest neighbor algorithm-based flotation operating condition recognition method is characterized by, In the step of weighting the K-Nearest Neighbor algorithm by the calculated characteristic function value, the current candidate characteristic The value of the target evaluation function is used as a standard to define the weight of the current candidate characteristic After the weight of each ranked characteristic is calculated, the Euclidean distance in the K-Nearest Neighbor algorithm is weighted. the current candidate feature a weight of: wherein, is the weight of the current candidate feature , is the value of the target evaluation function for the current candidate feature , is the total number of features; The weighting formula is: wherein, is the total number of features, and are the values of the features and the features of the sample respectively.
3. A flotation operating condition recognition system based on a weighted K- nearest neighbor algorithm, characterized by It comprises: A setting module (10) is configured to set fuzzy multi-neighborhood particles for fuzzy and uncertain data in a foam image dataset and different distributions of various features, and the setting module (10) comprises: An extraction unit (12) is configured to extract relevant features from each foam image sample in the sample set to form a feature set; A preprocessing unit (13) is configured to normalize the feature set to eliminate the dimension influence between different features; A setting unit (14) is configured to set fuzzy multi-neighborhood particles for various features after eliminating the dimension influence; A first calculation module (20) is configured to introduce the concept of entropy to calculate the information provided by the fuzzy multi-neighborhood particles induced by the feature subset, and fully consider the multiple relationships between the foam image features under the fuzzy multi-granularity information theory framework, including the correlation, redundancy, complementarity and interaction, and the first calculation module (20) comprises: a first computing unit (21) for computing the amount of information provided by the fuzzy multi-neighborhood granules induced by the feature subsets, said amount of information comprising the fuzzy multi-neighborhood mutual information of the sample set with respect to the feature subset A, the feature subset B, and the fuzzy conditional mutual information of the sample set given the feature subset A, the feature subset B and the feature subset C; The construction unit (22) is used for constructing a feature target evaluation function to measure the contribution degree of the froth image features to the image classification by fully considering the multiple relationships between the froth image features in the fuzzy multi-granularity information theory framework, and performing feature sorting, and the feature firstly entering the selected feature subset S is the feature with the largest correlation with the category D, and then the features with the largest feature target evaluation function are sequentially selected into the selected feature subset S until the features in the remaining candidate subset V are selected, and the calculation formula of the feature target evaluation function is as follows: wherein, relevance of the current candidate feature to the class, relevance of the current candidate feature to the selected feature subset, redundancy of the current candidate feature to the selected feature subset, complementarity of the current candidate feature to the selected feature, interaction of the remaining unselected features with the current candidate feature, interaction of the current candidate feature with the selected feature subset, the set of features, the selected feature subset, the remaining candidate feature subset; The second calculation module (30) is used for constructing a target evaluation function according to the multiple relationships between the features, calculating the function value of each feature, and sorting the features; The weighting module (40) is used for weighting the K nearest neighbor algorithm by using the calculated feature function value; The selection module (50) is used for selecting the feature subset with the highest classification accuracy as the optimal feature subset for the flotation condition recognition.
4. The floating operating condition identification system based on a weighted K- nearest neighbor algorithm as claimed in claim 3, characterized in that, The weighting module (40) weights the Euclidean distance in the K nearest neighbor algorithm according to the weight of each ranked feature. The value of the target evaluation function is used as a criterion to define the weight of the current candidate feature After the weight of each ranked feature is obtained, the Euclidean distance in the K nearest neighbor algorithm is weighted. the current candidate feature a weight of: in, For current candidate features The weight, For current candidate features The value of the objective evaluation function, The total number of features; The weighting formula is as follows: wherein, is the total number of features, and are the values of the features and the features of the sample respectively.