Mineral flotation working condition identification and prediction equipment and method

Abstract

The invention provides mineral flotation working condition identification and prediction equipment which comprises an image acquisition module, an ultrasonic ranging sensor, a main control module, a 5G communication module, a data transmission and positioning module, a display module and a power module. An ultrasonic ranging sensor above the flotation tank is used for measuring the height of the liquid level, the recognition algorithm is deployed on a helmet of a worker, the current working condition can be recognized in real time in the working process, the working trend of mineral flotation can be predicted, and the result is displayed on wearable recognition equipment. After wearing the equipment, a worker can predict fluctuation in the flotation process and master the operation state of a flotation tank, corresponding measures are taken in advance, the stability and the production efficiency of the mineral flotation process are further improved, and the safety risk and the economic loss are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mineral flotation, and in particular to a device and method for identifying and predicting mineral flotation conditions. Background Art

[0002] The mineral flotation process adds organic solvents to make the concentrate particles in the slurry adhere to the foam surface and be separated. Due to the low speed of the flotation tank's agitator, the slurry contains a large number of large particles, the motor failure causes the slurry to flow and accumulate, and the improper use of frothers, abnormal working conditions such as sinking and bubbling often occur, resulting in overload of flotation equipment, explosion or fire, chemical leakage endangering the health of workers, and a significant reduction in production efficiency. Safety and economic problems, therefore, need to be identified in a timely manner for mineral flotation conditions to reduce economic losses and improve production efficiency. The existing working condition identification methods are usually judged based on the experience and knowledge of professional technicians, or identified by measuring physical indicators such as the liquid level height and liquid pH value of the flotation tank. Only when the liquid level height and pH value in the flotation tank exceed the set threshold can the current abnormal working condition be identified and warned.

[0003] The traditional method of judging flotation conditions based on empirical knowledge has significant limitations and is accompanied by high labor costs. Due to individual differences in the experience of inspectors in judging flotation conditions, it often leads to inconsistency and subjective bias in the results of flotation condition judgment, and there is a high risk of misjudgment or missed judgment. At the same time, this method that relies on manual monitoring is not only inefficient, but also unreliable, and it is difficult to meet the requirements of modern industry for accurate detection.

[0004] The method of identifying working conditions by setting thresholds of physical indicators often faces challenges when faced with highly time-varying and dynamically complex flotation processes. Due to the volatility of flotation conditions, thresholds are difficult to set accurately. If the threshold is set too low, it may cause frequent false warnings and waste human resources; if it is set too high, it may cause warnings too late, resulting in significant economic losses. In addition, although the emerging working condition identification method based on image data can improve the detection accuracy of the current flotation conditions, it is still difficult to predict the trend of mineral flotation conditions.

[0005] Therefore, the working condition recognition algorithm based on multimodal data fusion of image data and liquid level information data is becoming a very promising direction in the field of mineral flotation working condition recognition and prediction. Summary of the invention

[0006] In order to make up for the deficiencies of the prior art, the embodiments of the present application propose a method for identifying and predicting mineral flotation conditions to solve the problems existing in the prior art.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: A mineral flotation condition identification and prediction device comprises an image acquisition module, an ultrasonic ranging sensor, a main control module, a 5G communication module, a data transmission and positioning module, a display module and a power module. The image acquisition module is responsible for acquiring image information of the current flotation condition. The ultrasonic ranging sensor above the flotation tank is used to measure the height of the liquid surface. The image and height data are respectively transmitted to the main control module via the MIPI CSI interface in the data transmission and positioning module and the built-in Bluetooth module. The 5G communication module is used to realize real-time data upload and remote monitoring between the flotation device and the remote server. The main control module applies a condition identification and prediction algorithm to identify and predict the trend of the current flotation condition. The identification and prediction results are presented via VR glasses in the display module.

[0008] A method for identifying and predicting mineral flotation conditions, using the device described in claim 1, comprising the following steps: Step 1: Construct a mineral flotation condition dataset. Collect various condition images through the actual flotation process on industrial site and a self-made mineral flotation condition recognition platform. Re-classify the labels into sink, normal and bubbling tank. Then divide the self-made condition dataset into a validation set and a training set. Step 2: Build a model. Based on the improved YOLOV8, build the mineral flotation condition recognition model SD-LSTM-Model. First, use SD-Darknet as the backbone network to extract features, including: strided convolution layer, SD-C2f cross-stage feature fusion layer, SD-SPPF pooling layer, fully connected layer and activation function. Then, use the neck network to fuse and enhance the features. Finally, call three decoupled head detectors and prediction modules for detection and prediction. Step 3: Feature extraction: Use the SD-C2f module to connect feature maps at different levels, extract dense features at different scales, focus on more important channel features, and then use the SD-SPPF fast spatial pyramid pooling layer to extract deep information from the spatial dimension and channel dimension respectively, so as to improve the network's ability to extract detailed information from photos of mineral flotation conditions; Step 4: Feature fusion: extract the spatial features of the image data in the YOLO detection module, fuse the spatial features with the time series measured by the ultrasonic distance sensor through the self-attention mechanism method, and send them into the LSTM network to predict the mineral flotation conditions at the next moment; Step 5: Detection and prediction: input the working condition photos taken by the equipment camera and the data obtained by the ultrasonic rangefinder into the detector to identify the working condition and predict the working trend of mineral flotation so that the staff can take countermeasures in advance.

[0009] As a further technical solution of the present invention: constructing a data set specifically includes: collecting samples required for model training based on a self-made mineral flotation condition identification platform, and using Label Img to redefine labels as settling tank, normal and bubbling tank according to different working conditions, with a total of 50,000 labels. The redefined labels and samples constitute a new mineral flotation condition identification data set.

[0010] As a further technical solution of the present invention: the model building is specifically: building a deep convolutional neural network model SD-LSTM-Model based on Yolov8 improvement, including: a deep convolutional network SD-Darknet, an SD-C2f module, an SD-SPPF module, a feature fusion module and a decoupling head detector, and a prediction module based on LSTM. When a picture is input into the model, the backbone network extracts the feature information of the picture through strided convolution. The SD-C2f module integrates the channel attention mechanism of SE, so that the neural network pays more attention to important channel features, and then extracts the depth detail information through the SD-SPPF module. Through feature fusion and enhancement, the decoupling detection head is used to detect the video, the recognition result is output, the spatial image features in the detection result are extracted, and the liquid level height data collected by the ultrasonic sensor is integrated and input into the LSTM network for prediction.

[0011] As a further technical solution of the present invention: the feature extraction: first, through a series of standard convolutions, the basic features of the image are extracted, based on the improvement of the C2f module of Darknet-53, the SD-C2f module is designed, the input feature map is channel-segmented, and each channel is processed separately to improve the distinction between features, then, the feature map after channel separation is processed by N Bottleneck2 modules, the processed feature map is spliced ​​with the feature map after other channel separation to obtain the connection between different channel features, and then processed by Bottlencek2 modules in turn to reduce the number of parameters and speed up the calculation, the above feature extraction process is repeated N times, the N processed feature maps are spliced, and the SE attention mechanism module is introduced to adaptively adjust and proofread the feature weights, so that the network can dynamically adjust the attention of the channel and strengthen the utilization of effective information, and finally the feature map is processed by the residual network to obtain feature information rich in the gradient flow and avoid the problems of gradient explosion and network performance degradation.

[0012] As a further technical solution of the present invention: the SE channel attention mechanism module includes Squeeze and Excitation operations, which are intended to make the neural network pay more attention to the important features of the current task. First, the global spatial information is compressed into a channel descriptor, that is, a global flat pooling operation is performed on each channel, and the features containing the global information with a dimension of H*W*C are compressed into a feature vector Z of 1*1*C. The channel features of C feature maps are compressed into a numerical value, and a channel-level statistical data Z with contextual information is generated to solve the problem of channel dependency. Z C is defined as follows: (1) Among them, Z C is the Cth element of Z, F sq Refers to the global average pooling of channels, U C are the C output feature maps obtained; In order to utilize the information gathered in the compression operation, the Excitation excitation operation is used to fully capture the dependencies between channels. A two-layer fully connected gate mechanism is adopted. The first fully connected layer reduces the C channels of the feature map to C / m channels to reduce the amount of calculation. After processing through a ReLU nonlinear activation layer, it is input into the second fully connected layer to restore the number of channels to C channels, and the weight P is obtained through Sigmoid activation. Finally, P with 1*1*C dimensions is obtained, which is used to represent the weights of C feature maps in the feature map U, where m refers to the compression ratio. P is expressed as follows: (2) Finally, in order to weight the obtained attention weight P to the features of each channel, it is necessary to multiply each feature map in the feature map U by the corresponding weight to obtain the final output T of the SE attention mechanism module, which is expressed as follows: (3) The feature map extracted by the SD-Darknet backbone network is input into the SD-SPPF spatial feature pyramid pooling layer. The role of SPPF spatial pyramid pooling is to apply the maximum pooling operation of different scales to the entire feature map. First, the input feature map is divided into different grid areas, and then the maximum pooling operation is applied to each area to generate a fixed-length feature vector. Finally, the feature vectors generated at these different scales are spliced ​​together to form a fixed-length feature vector to meet the requirements of fixed-dimensional input when the fully connected layer implements classification. The SD-SPPF module first uses DW depth-wise separable convolution to extract features, and then forms a feature vector of fixed length through maximum pooling and splicing. DW convolution consists of depth-wise convolution and point-wise convolution. In depth-wise convolution, the input channel and a separate filter are first convolved to obtain a feature map consistent with the number of channels, thereby realizing the extraction of spatial features. In point-by-point convolution, a 1x1 convolution kernel is used to increase the dimension of the output of the depth-wise convolution and compress it to the required number of channels to extract channel features. Assume that the convolution kernel size is D K *D K , the input channel is M, the output channel is N, and the output feature map size is D F *D F ,The comparison value of the depth-separable convolution and the standard convolution parameter λ: (4) Generally, N is large and can be ignored. When we use the most common 3*3 convolution kernel, the number of parameters and computational complexity of the depthwise separable convolution can be reduced to one-ninth of the original, making it easier to deploy on a helmet device for working condition identification.

[0013] As a further technical solution of the present invention: the feature fusion is specifically: using the SD-LSTM-Model model to extract the spatial features of the current liquid level, namely the liquid level height, from the image, capturing the relationship between the liquid level height and the spatial features, and obtaining the fused feature representation.

[0014] As a further technical solution of the present invention: the detection and prediction are specifically: taking the working condition photos taken by the camera of the identification equipment and the liquid level height data obtained by the ultrasonic rangefinder as input, the current working condition of mineral flotation is identified and predicted through the designed SD-LSTM-Model network, and the results are displayed on the VR glasses of the homemade flotation working condition identification and trend prediction equipment.

[0015] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: By deploying the recognition algorithm on the helmet of the worker, it is possible to identify the current working conditions in real time during the work process and predict the working trend of mineral flotation, and display the results on the wearable recognition device. After wearing the device, the worker can foresee the fluctuations in the flotation process and grasp the operating status of the flotation tank, take corresponding measures in advance, further improve the stability and production efficiency of the mineral flotation process, reduce safety risks and reduce economic losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is the structural diagram of the flotation condition identification and trend prediction equipment.

[0017] Figure 2 It is a flow chart for flotation condition identification and prediction.

[0018] Figure 3 This is the SD-LSTM-Model structure diagram.

[0019] Figure 4 This is the SD-C2f structure diagram.

[0020] Figure 5 This is the SD-SPPF structure diagram.

[0021] Figure 6 This is the result chart of the homemade detection system. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Reference Figure 1-6 , a mineral flotation condition identification and prediction device, consists of a series of modules: image acquisition module, ultrasonic ranging sensor, main control module, 5G communication module, data transmission and positioning module, display module (VR glasses), power module; Among them, the image acquisition module is responsible for obtaining the image information of the current flotation condition, while the ultrasonic ranging sensor above the flotation tank is used to measure the height of the liquid surface. The image and height data are transmitted to the main control module through the MIPI CSI interface in the data transmission and positioning module and the built-in Bluetooth module respectively to ensure efficient data transmission and precise positioning of the equipment. At the same time, the 5G communication module realizes real-time data upload and remote monitoring between the flotation equipment and the remote server. The main control module then applies the working condition identification and prediction algorithm to identify and predict the trend of the current flotation condition. Finally, the identification and prediction results are presented through the VR glasses in the display module to help operators grasp the working conditions and their development trends in real time. The structure of the flotation condition identification and trend prediction equipment is as follows: Figure 1 shown.

[0024] At the same time, the present invention also provides a method for identifying and predicting mineral flotation conditions, comprising the following steps: Step 1: Construct a mineral flotation condition dataset. Collect various working condition pictures through the actual flotation process on industrial site and a self-made mineral flotation condition identification platform. Re-classify the labels into settling tank, normal and bubbling tank. Then divide the self-made working condition dataset into a validation set and a training set.

[0025] Step 2, build the model: build the mineral flotation condition recognition model SD-LSTM-Model based on the improved YOLOV8. First, use SD-Darknet as the backbone network to extract features, including: strided convolution layer, SD-C2f cross-stage feature fusion layer, SD-SPPF pooling layer, fully connected layer and activation function, etc. Then, the features are fused and enhanced through the neck network, and finally three decoupled head detectors and prediction modules are called for detection and prediction.

[0026] Step 3, feature extraction: Use the SD-C2f module to connect feature maps at different levels, extract dense features at different scales, and focus on more important channel features. Then use the SD-SPPF fast spatial pyramid pooling layer to extract deep information from the spatial dimension and channel dimension respectively, and improve the network's ability to extract detailed information from photos of mineral flotation conditions.

[0027] Step 4, feature fusion stage: extract the spatial features of the image data in the YOLO detection module, fuse the spatial features with the time series measured by the ultrasonic distance sensor through methods such as the self-attention mechanism, and send them into the LSTM network to predict the mineral flotation conditions at the next moment.

[0028] Step 5, detection and prediction stage: The working condition photos taken by the equipment camera and the data obtained by the ultrasonic rangefinder are input into the detector to identify the working condition and predict the working trend of mineral flotation, so that the staff can take countermeasures in advance.

[0029] (1) Constructing the data set: Based on the self-made mineral flotation condition identification platform, the samples required for model training were collected. According to different working conditions, the labels were redefined as sink, normal and bubbling tank using Label Img, with a total of 50,000 labels. The redefined labels and samples constitute a new mineral flotation condition identification data set.

[0030] (2) Model building: Building a deep convolutional neural network model SD-LSTM-Model based on Yolov8 ( Figure 3 ), including: deep convolutional network SD-Darknet, SD-C2f module ( Figure 4 )、SD-SPPF module( Figure 5), feature fusion module and decoupling head detector, and LSTM-based prediction module. When inputting an image into the model, the backbone network extracts the feature information of the image through strided convolution. The SD-C2f module integrates the channel attention mechanism of SE to make the neural network pay more attention to important channel features. The SD-SPPF module then extracts the depth detail information, and through feature fusion and enhancement, the decoupling detection head is used to detect the video and output the recognition result. The spatial image features in the detection results are extracted, the liquid level data collected by the ultrasonic sensor is integrated, and the data is input into the LSTM network for prediction.

[0031] (3) Feature extraction: First, a series of standard convolutions are used to extract the basic features of the image. Based on the improvement of the C2f module of Darknet-53, the SD-C2f module is designed. The input feature map is divided into channels to process each channel separately to improve the discrimination between features. Then, N Bottleneck2 modules are used to process the feature map after channel separation, and the processed feature map is spliced ​​with the feature map after other channel separation to obtain the connection between the features of different channels, and then processed by Bottlencek2 modules in turn to reduce the number of parameters and speed up the calculation. The above feature extraction process is repeated N times, and the N processed feature maps are spliced. The SE attention mechanism module is introduced to adaptively adjust and proofread the feature weights, so that the network can dynamically adjust the attention of the channel and enhance the use of effective information. Finally, the feature map is processed by the residual network to obtain feature information that enriches the gradient flow and avoids the problems of gradient explosion and network performance degradation.

[0032] The SE channel attention mechanism module mainly includes Squeeze and Excitation operations, which are designed to make the neural network pay more attention to the important features of the current task. First, the global spatial information is compressed into a channel descriptor, that is, a global flat pooling operation is performed on each channel, and the features with dimensions H*W*C containing global information are compressed into a 1*1*C feature vector Z. The channel features of C feature maps are compressed into a numerical value, and channel-level statistics Z with contextual information are generated to solve the problem of channel dependency. C is defined as follows: (1) Among them, Z C is the Cth element of Z, F sq Refers to the global average pooling of channels, U C These are the C output feature maps obtained.

[0033] In order to utilize the information gathered in the compression operation, the Excitation excitation operation is used to fully capture the dependencies between channels. A two-layer fully connected gate mechanism is used. The first fully connected layer reduces the C channels of the feature map to C / m channels to reduce the amount of calculation. After processing through a ReLU nonlinear activation layer, it is input into the second fully connected layer to restore the number of channels to C channels, and the weight P is obtained through Sigmoid activation. Finally, P of 1*1*C dimensions is obtained to represent the weights of C feature maps in feature map U, where m refers to the compression ratio. P is expressed as follows: (2) Finally, in order to weight the obtained attention weight P to the features of each channel, it is necessary to multiply each feature map in the feature map U by the corresponding weight to obtain the final output T of the SE attention mechanism module. T is expressed as follows: (3) The feature map extracted by the SD-Darknet backbone network is input into the SD-SPPF spatial feature pyramid pooling layer. The role of SPPF spatial pyramid pooling is to apply the maximum pooling operation of different scales to the entire feature map. First, the input feature map is divided into different grid areas, and then the maximum pooling operation is applied to each area to generate a fixed-length feature vector. Finally, the feature vectors generated at these different scales are spliced ​​together to form a fixed-length feature vector to meet the fixed-dimensional input requirements when the fully connected layer implements classification.

[0034] The SD-SPPF module first uses DW depth-wise separable convolution to extract features, and then forms a fixed-length feature vector through maximum pooling and concatenation. DW convolution consists of depth-wise convolution and point-wise convolution. In depth-wise convolution, the input channel and a separate filter are first convolved to obtain a feature map consistent with the number of channels, thereby realizing the extraction of spatial features.

[0035] In point-by-point convolution, a 1x1 convolution kernel is used to increase the dimension of the output of depth-wise convolution and compress it to the required number of channels to extract channel features. Assume that the convolution kernel size is D K *D K , the input channel is M, the output channel is N, and the output feature map size is D F *D F ,The comparison value of the depth-separable convolution and the standard convolution parameter λ: (4) Generally, N is large and can be ignored. When we use the most common 3*3 convolution kernel, the number of parameters and computational complexity of depthwise separable convolution can be reduced to one-ninth of the original, making it easier to deploy on helmet devices for working condition recognition.

[0036] (4) Feature fusion stage: Use the SD-LSTM-Model model to extract the spatial features of the current liquid surface (including liquid level, area, etc.) from the image, effectively capture the relationship between the liquid level and spatial features, and obtain the fused feature representation.

[0037] First, the liquid level data of the sensor is mapped to the dimensional space of the query vector through the fully connected layer through the self-attention mechanism to generate the query vector Q (Query). The SD-LSTM-Model model is used to extract the spatial features of the current liquid level (including liquid level, area, etc.) from the image, and generate the key vector K (Key) representing the importance of the image features and the value vector V (Value) representing the content of the image spatial features. The dot product is used to calculate the similarity between the query vector and the key vector to generate the attention score matrix, and then the score matrix is ​​normalized to convert the score matrix into a weight matrix to represent the importance of each spatial feature to the sensor data feature. The value vector is weighted and summed using the attention weight to obtain the fusion vector 'attention_features' that contains both image features and sensor features weighted by relevance.

[0038] At the same time, the liquid level features of the sensor are extracted through the sliding window, and horizontally connected with the spatial features of the liquid surface to form a spliced ​​fusion feature vector 'sum_features'. The above two fusion feature vectors are processed by Transform-Concatenate-Transform. First, the 'attention_features' and 'sum_features' are linearly transformed through the fully connected layer, and then the transformed feature vectors are spliced ​​to form a new fusion feature vector 'combined_features'. The fused feature vector is then linearly transformed through the fully connected layer to capture the relationship between the liquid level and the spatial features, and finally the fused feature representation is obtained.

[0039] The fused features are preprocessed and converted into a three-dimensional tensor suitable for input into the LSTM network to train the LSTM model and predict the change in the liquid level at the next moment. t Positive (liquid level rises), negative (liquid level drops), and the results are displayed on the glasses that identify the helmet.

[0040] (5) Detection and prediction stage: Using the working condition photos taken by the recognition device camera and the liquid level height data obtained by the ultrasonic rangefinder as input, the designed SD-LSTM-Model network is used to identify and predict the current working condition of the mineral flotation, and the results are displayed on the VR glasses of the homemade flotation condition identification and trend prediction device.

[0041] In view of the difficulty of realizing accurate identification and prediction of flotation conditions by single factors such as physical indicators or image information, the present invention integrates the liquid level and flotation condition image information, constructs a mineral flotation condition identification and prediction model SD-LSTM-Model, and deploys the model on a wearable device. By performing multimodal data fusion enhancement on image data and liquid level information, the LSTM network is used to learn the change of liquid level in the flotation tank, and the working trend of mineral flotation is predicted based on the height change results. The working condition detection results and trends are displayed on the VR glasses of the wearable device, which is convenient for the staff to observe in real time, so as to make corresponding prevention and adjustment in advance.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

[0043] In addition, it should be understood that although this specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment have also been appropriately combined to form other implementation modes that are easy for those skilled in the art to understand.

Claims

1. A mineral flotation condition identification and prediction device, comprising an image acquisition module, an ultrasonic ranging sensor, a main control module, a 5G communication module, a data transmission and positioning module, a display module and a power module, characterized in that: The image acquisition module is responsible for acquiring image information of the current flotation condition. The ultrasonic ranging sensor above the flotation tank is used to measure the height of the liquid level. The image and height data are transmitted to the main control module through the MIPI CSI interface in the data transmission and positioning module and the built-in Bluetooth module respectively. The 5G communication module is used to realize real-time data upload and remote monitoring between the flotation equipment and the remote server. The main control module applies the working condition recognition and prediction algorithm to identify and predict the trend of the current flotation condition. The recognition and prediction results are presented through the VR glasses in the display module.

2. A method for identifying and predicting mineral flotation conditions, characterized in that: The device according to claim 1 comprises the following steps: Step 1: Construct a mineral flotation condition dataset. Collect various condition images through the actual flotation process on industrial site and a self-made mineral flotation condition recognition platform. Re-classify the labels into sink, normal and bubbling tank. Then divide the self-made condition dataset into a validation set and a training set. Step 2: Build a model. Based on the improved YOLOV8, build the mineral flotation condition recognition model SD-LSTM-Model. First, use SD-Darknet as the backbone network to extract features, including: strided convolution layer, SD-C2f cross-stage feature fusion layer, SD-SPPF pooling layer, fully connected layer and activation function. Then, use the neck network to fuse and enhance the features. Finally, call three decoupled head detectors and prediction modules for detection and prediction. Step 3: Feature extraction: Use the SD-C2f module to connect feature maps at different levels, extract dense features at different scales, focus on more important channel features, and then use the SD-SPPF fast spatial pyramid pooling layer to extract deep information from the spatial dimension and channel dimension respectively, so as to improve the network's ability to extract detailed information from photos of mineral flotation conditions; Step 4: Feature fusion: extract the spatial features of the image data in the YOLO detection module, fuse the spatial features with the time series measured by the ultrasonic distance sensor through the self-attention mechanism method, and send them into the LSTM network to predict the mineral flotation conditions at the next moment; Step 5: Detection and prediction: input the working condition photos taken by the equipment camera and the data obtained by the ultrasonic rangefinder into the detector to identify the working condition and predict the working trend of mineral flotation so that the staff can take countermeasures in advance.

3. A method for identifying and predicting mineral flotation conditions according to claim 2, characterized in that: The specific steps of constructing the data set are as follows: based on the self-made mineral flotation condition identification platform, samples required for model training are collected. According to different working conditions, Label Img is used to redefine labels as settling tank, normal and bubbling tank, with a total of 50,000 labels. The redefined labels and samples constitute a new mineral flotation condition identification data set.

4. A method for identifying and predicting mineral flotation conditions according to claim 2, characterized in that: The model building is specifically: building a deep convolutional neural network model SD-LSTM-Model based on Yolov8 improvement, including: deep convolutional network SD-Darknet, SD-C2f module, SD-SPPF module, feature fusion module and decoupling head detector, LSTM-based prediction module. When inputting a picture into the model, the backbone network extracts the feature information of the picture through strided convolution. The SD-C2f module integrates the channel attention mechanism of SE to make the neural network pay more attention to important channel features. The SD-SPPF module is used to extract the depth detail information. Through feature fusion and enhancement, the decoupling detection head is used to detect the video, the recognition result is output, the spatial image feature in the detection result is extracted, the liquid level height data collected by the ultrasonic sensor is integrated, and the data is input into the LSTM network for prediction.

5. A method for identifying and predicting mineral flotation conditions according to claim 2, characterized in that: The feature extraction is as follows: first, a series of standard convolutions are used to extract the basic features of the image. Based on the improvement of the C2f module of Darknet-53, an SD-C2f module is designed to divide the input feature map into channels so as to process each channel separately to improve the distinction between features. Then, N Bottleneck2 modules are used to process the feature map after channel separation, and the processed feature map is spliced ​​with the feature map after other channel separation to obtain the connection between different channel features. Then, the Bottlencek2 modules are used in turn to reduce the number of parameters and speed up the calculation. The above feature extraction process is repeated N times, and the N processed feature maps are spliced. The SE attention mechanism module is introduced to adaptively adjust and proofread the feature weights, so that the network can dynamically adjust the attention of the channel and strengthen the use of effective information. Finally, the residual network is used to process the feature map to obtain feature information rich in the gradient flow and avoid the problems of gradient explosion and network performance degradation.

6. A method for identifying and predicting mineral flotation conditions according to claim 5, characterized in that: The SE channel attention mechanism module includes Squeeze and Excitation operations, which are designed to make the neural network pay more attention to the important features of the current task. First, the global spatial information is compressed into a channel descriptor, that is, a global flat pooling operation is performed on each channel, and the features with dimensions H*W*C containing global information are compressed into a feature vector Z of 1*1*C. The channel features of C feature maps are compressed into a numerical value, and channel-level statistics Z with contextual information are generated to solve the problem of channel dependency. C is defined as follows: (1) Among them, Z C is the Cth element of Z, F sq Refers to the global average pooling of channels, U C are the C output feature maps obtained; In order to utilize the information gathered in the compression operation, the Excitation excitation operation is used to fully capture the dependencies between channels. A two-layer fully connected gate mechanism is adopted. The first fully connected layer reduces the C channels of the feature map to C / m channels to reduce the amount of calculation. After processing through a ReLU nonlinear activation layer, it is input into the second fully connected layer to restore the number of channels to C channels, and the weight P is obtained through Sigmoid activation. Finally, P with 1*1*C dimensions is obtained, which is used to represent the weights of C feature maps in the feature map U, where m refers to the compression ratio. P is expressed as follows: (2) Finally, in order to weight the obtained attention weight P to the features of each channel, it is necessary to multiply each feature map in the feature map U by the corresponding weight to obtain the final output T of the SE attention mechanism module, which is expressed as follows: (3) The feature map extracted by the SD-Darknet backbone network is input into the SD-SPPF spatial feature pyramid pooling layer. The role of SPPF spatial pyramid pooling is to apply the maximum pooling operation of different scales to the entire feature map. First, the input feature map is divided into different grid areas, and then the maximum pooling operation is applied to each area to generate a fixed-length feature vector. Finally, the feature vectors generated at these different scales are spliced ​​together to form a fixed-length feature vector to meet the requirements of fixed-dimensional input when the fully connected layer implements classification. The SD-SPPF module first uses DW depth-wise separable convolution to extract features, and then forms a feature vector of fixed length through maximum pooling and splicing. DW convolution consists of depth-wise convolution and point-wise convolution. In depth-wise convolution, the input channel and a separate filter are first convolved to obtain a feature map consistent with the number of channels, thereby realizing the extraction of spatial features. In point-by-point convolution, a 1x1 convolution kernel is used to increase the dimension of the output of the depth-wise convolution and compress it to the required number of channels to extract channel features. Assume that the convolution kernel size is D K *D K , the input channel is M, the output channel is N, and the output feature map size is D F *D F ,The comparison value of the depth-separable convolution and the standard convolution parameter λ: （4） Generally, N is large and can be ignored. When we use the most common 3*3 convolution kernel, the number of parameters and computational complexity of the depthwise separable convolution can be reduced to one-ninth of the original, making it easier to deploy on a helmet device for working condition identification.

7. A method for identifying and predicting mineral flotation conditions according to claim 2, characterized in that: The feature fusion specifically includes: using the SD-LSTM-Model model to extract the spatial features of the current liquid level, namely the liquid level height, from the image, capturing the relationship between the liquid level height and the spatial features, and obtaining the fused feature representation.

8. A method for identifying and predicting mineral flotation conditions according to claim 2, characterized in that: The detection and prediction are specifically as follows: using the working condition photos taken by the recognition device camera and the liquid level height data obtained by the ultrasonic rangefinder as input, the designed SD-LSTM-Model network is used to identify and predict the current working condition of mineral flotation, and the results are displayed on the VR glasses of the homemade flotation working condition identification and trend prediction device.

Images