Lithium mineral microscopic image segmentation method and device

Abstract

The invention provides a lithium mineral microscopic image segmentation method and device, and relates to the technical field of computer vision and image segmentation, and the method comprises the steps: obtaining lithium mineral microscopic image data, and generating labeled sample data; according to the annotation sample data, training a preset initial version YOLO data annotation segmentation model; based on a preset initial version YOLO data annotation segmentation model, generating a preset advanced version YOLO data annotation segmentation model, and training the preset advanced version YOLO data annotation segmentation model; and performing lithium mineral microscopic image segmentation by using the trained advanced YOLO data annotation segmentation model. According to the embodiment of the invention, the segmentation model is marked by using the advanced YOLO data, so that the calculation efficiency of the model is balanced while the detection precision and the segmentation precision are ensured. The model can achieve high precision in practical application, has enough real-time processing capability, and adapts to application requirements in a dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision and image segmentation, in particular to a lithium mineral microscopic image segmentation method and device. BACKGROUND

[0002] The accurate identification of lithium ore mineral quality directly determines the economic benefits and industrial application value of intelligent mining. Under the background of the accelerated transformation of intelligent mining, the development of mineral identification technology with real-time analysis capability not only concerns the improvement of resource development efficiency, but also is a key link to realize green and sustainable mining. Efficient mineral grade identification technology can not only obtain real-time ore quality data, but also optimize production decision-making processes and effectively avoid resource mismatch and environmental pollution risks.

[0003] However, traditional lithium ore quality detection methods mainly rely on manual sampling and laboratory analysis. This method has limitations such as long detection cycle, poor real-time performance, and susceptibility to subjective experience interference, making it difficult to meet the needs of modern mines for rapid and accurate identification of ore quality. With the rapid development of computer vision technology, deep learning-based technology has been gradually applied to mineral composition analysis.

[0004] In contrast, deep learning algorithms, as a leading technology in intelligent mineral segmentation, break through the limitations of complex instrument operation and manual segmentation modeling in traditional methods by automatically mining the complex relationship between optical features and mineral grade. Relying on non-contact visual detection, it realizes online analysis in seconds, providing a more efficient solution for industrial continuous detection. However, existing deep learning-based lithium mineral quality segmentation techniques still face some common shortcomings. For example, CN119478398A proposes using a SAM segmentation model to segment lithium mineral microscopic images and implementing automatic segmentation using manually labeled data. However, this method has the following obvious drawbacks: the method has poor real-time performance and cannot meet the needs of field applications. SAM and other segmentation models generally have large computational load and high complexity, and the image segmentation process may take a long time to calculate, affecting real-time performance. In mining, metallurgy, and other industrial fields, real-time evaluation of lithium mineral quality is crucial, and fast inference speed is essential for optimizing production decisions. Therefore, this poor real-time performance problem restricts its application in actual production.

[0005] In view of the deficiencies of the prior art, there is an urgent need for a mineral segmentation method with higher real-time performance. SUMMARY

[0006] To solve the technical problem of low real-time performance of the existing mineral segmentation method, the present application provides a lithium mineral microscopic image segmentation method and device. The technical solution is as follows:

[0007] In one aspect, a lithium mineral microscopic image segmentation method is provided, comprising:

[0008] acquiring lithium mineral microscopic image data, performing feature extraction on the lithium mineral microscopic image data;

[0009] performing data optimization on the lithium mineral microscopic image data after feature extraction, dividing the optimized lithium mineral microscopic image data into first part data and second part data, and manually labeling the first part data to generate labeled sample data;

[0010] training a preset initial version YOLO data labeling and segmentation model according to the labeled sample data, and labeling the second part data using the trained initial version YOLO data labeling and segmentation model;

[0011] based on the preset initial version YOLO data labeling and segmentation model, generating a preset advanced version YOLO data labeling and segmentation model, dividing the labeled sample data and the labeled second part data into a validation set and a training set, training the preset advanced version YOLO data labeling and segmentation model using the validation set and the training set, and obtaining a trained advanced version YOLO data labeling and segmentation model;

[0012] segmenting the lithium mineral microscopic image using the trained advanced version YOLO data labeling and segmentation model.

[0013] Optionally, the generating of the preset advanced version YOLO data labeling and segmentation model based on the preset initial version YOLO data labeling and segmentation model comprises:

[0014] scaling the channel number of the backbone network part of the preset initial version YOLO data labeling and segmentation model by equal proportion, and reducing the channel number of each backbone layer by half;

[0015] replacing all conventional convolutions of the backbone network and the neck network of the preset initial version YOLO data labeling and segmentation model with lightweight windmill-shaped convolutions;

[0016] removing all channel-by-channel convolutions in the head network of the preset initial version YOLO data labeling and segmentation model, and constructing a lightweight Segment-Efficient segmentation head.

[0017] Optionally, the training of the preset advanced version YOLO data labeling and segmentation model using the validation set and the training set comprises:

[0018] performing initialization of various parameters of the preset advanced version YOLO data labeling and segmentation model, wherein the various parameters include the total number of training rounds, the number of samples processed per round, the patience threshold, and the initial learning rate;

[0019] The advanced YOLO data labeling segmentation model is trained using the training set, and after each training period, the performance of the advanced YOLO data labeling segmentation model is monitored using the verification set; in the case that the accuracy of the advanced YOLO data labeling segmentation model continues to improve and the loss value gradually decreases, the training is continued; in the case that the accuracy of the advanced YOLO data labeling segmentation model decreases or the loss value increases, the training is terminated.

[0020] Optionally, the feature extraction on the lithium mineral microscopic image data comprises:

[0021] The morphological features, surface texture distribution and particle boundary information of the lithium mineral microscopic image data are extracted.

[0022] Optionally, the data optimization on the lithium mineral microscopic image data comprises:

[0023] The small area removal and area filtering technology are used to set a filtering threshold by analyzing the area of the connected region in the lithium mineral microscopic image data, and the over-separated small debris in the lithium mineral microscopic image data is removed.

[0024] Optionally, the backbone network of the preset initial version YOLO data labeling segmentation model is responsible for extracting low-level features of the lithium mineral microscopic image data.

[0025] The neck network of the preset initial version YOLO data labeling segmentation model is responsible for aggregating and optimizing the low-level features of the lithium mineral microscopic image data to improve the detection ability of different scale targets.

[0026] The head network of the preset initial version YOLO data labeling segmentation model is responsible for generating the detection result of the target.

[0027] Optionally, the labeling of the second part of data using the trained initial version YOLO data labeling segmentation model comprises:

[0028] According to the trained target detection weight, the initial version YOLO data labeling segmentation model is used to detect the target of the second part of data, and the target coordinates are obtained.

[0029] The detection frame is generated according to the target coordinates.

[0030] According to the trained segmentation weight, the initial version YOLO data labeling segmentation model is used to generate a segmentation mask with a class label for the ore block image in the detection frame.

[0031] In another aspect, a lithium mineral microscopic image segmentation device is provided, which is used to implement the lithium mineral microscopic image segmentation method provided by the embodiments of the present application, and the device comprises:

[0032] An acquisition module is configured to acquire lithium mineral microscopic image data and perform feature extraction on the lithium mineral microscopic image data.

[0033] A labeling module is configured to perform data optimization on the feature-extracted lithium mineral microscopic image data, divide the optimized lithium mineral microscopic image data into first part data and second part data, and manually label the first part data to generate labeled sample data.

[0034] An initial version model training module is configured to train a preset initial version YOLO data labeling and segmentation model according to the labeled sample data, and label the second part data using the trained initial version YOLO data labeling and segmentation model.

[0035] An advanced version model training module is configured to generate a preset advanced version YOLO data labeling and segmentation model based on the preset initial version YOLO data labeling and segmentation model, divide the labeled sample data and the labeled second part data into a verification set and a training set, train the preset advanced version YOLO data labeling and segmentation model using the verification set and the training set, and obtain a trained advanced version YOLO data labeling and segmentation model.

[0036] A segmentation module is configured to segment lithium mineral microscopic images using the trained advanced version YOLO data labeling and segmentation model.

[0037] In another aspect, a lithium mineral microscopic image segmentation device is provided, and the lithium mineral microscopic image segmentation device includes:

[0038] A processor;

[0039] A memory having computer readable instructions stored thereon, the computer readable instructions being executed by the processor to implement the method provided by the embodiments of the present application.

[0040] In another aspect, a computer readable storage medium is provided, and the computer readable storage medium stores program codes, the program codes being called and executed by a processor to implement the method provided by the embodiments of the present application.

[0041] The technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0042] The embodiment of the present application can realize the following beneficial effects by segmenting the lithium mineral microscopic image through the advanced version YOLO data labeling segmentation model: (1) automatic labeling method: a small amount of data labeled by hand is used to train a detection model, and then the obtained target detection and segmentation pre-training weight is used to automatically label a large-scale data set, thereby effectively reducing the workload of a large number of instance labeling in the image segmentation task. (2) balance between precision and efficiency: by using the advanced version YOLO data labeling segmentation model, the calculation efficiency of the model is balanced while ensuring the detection precision and segmentation precision. The model can achieve high precision in practical application, and also has sufficient real-time processing capability, which adapts to the application requirements in a dynamic environment. (3) industrial application deployment: the advanced version YOLO data labeling segmentation model of the present application can be deployed in lithium mica ore flotation effect evaluation intelligent equipment, and supports real-time industrial field analysis and detection. The recognition efficiency of effective components of lithium minerals can be greatly improved, and intelligent support is provided for mining production and resource optimization. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] Figure 1 is a lithium mineral microscopic image segmentation method flowchart provided by the embodiment of the present application;

[0045] Figure 2 is a lithium mineral microscopic image data optimization schematic diagram provided by the embodiment of the present application;

[0046] Figure 3 is a structure schematic diagram of an initial version YOLO data labeling segmentation model provided by the embodiment of the present application;

[0047] Figure 4 is a method flowchart for generating an advanced version YOLO data labeling segmentation model provided by the embodiment of the present application;

[0048] Figure 5 is a structure schematic diagram of an advanced version YOLO data labeling segmentation model provided by the embodiment of the present application;

[0049] Figure 6 is a PSConv convolution module flowchart provided by the embodiment of the present application;

[0050] Figure 7 is a head structure schematic diagram of an initial version YOLO data labeling segmentation model provided by the embodiment of the present application;

[0051] Figure 8 is a Segment-Efficient structure schematic diagram provided by an embodiment of the present application;

[0052] Figure 9 is a training flowchart of an advanced version YOLO data labeling segmentation model provided by an embodiment of the present application;

[0053] Figure 10 is a method flowchart for labeling the second part of data provided by an embodiment of the present application;

[0054] Figure 11 is a segmentation mask precision and confidence curve provided by an embodiment of the present application;

[0055] Figure 12 is a segmentation effect example diagram provided by an embodiment of the present application

[0056] Figure 13 is a structure schematic diagram of a lithium mineral microscopic image segmentation device provided by an embodiment of the present application;

[0057] Figure 14 is a structure schematic diagram of a lithium mineral microscopic image segmentation device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the present application will be described below with reference to the drawings.

[0059] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0060] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0061] In the embodiments of the present application, sometimes the subscript such as W1 can be mistakenly used in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0062] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0063] In order to solve the technical problem of low real-time performance of the existing mineral separation method, the embodiments of the present application provide a lithium mineral microscopic image segmentation method and device. The technical solution is as follows:

[0064] As shown in Figure 1 The present application provides a lithium mineral microscopic image segmentation method, which comprises:

[0065] S1, obtaining lithium mineral microscopic image data, and extracting features from the lithium mineral microscopic image data.

[0066] After collecting finished lithium mineral powder in an industrial field, imaging is performed using a microscope, and the main configuration parameters are as follows: the resolution is 3840x2160 pixels, the objective lens is 0.5x (focal length: 0.7, magnification: 200x), and the object distance is 120mm. A large number of high-resolution microscopic image samples are obtained through this process, providing high-quality raw data for subsequent data optimization, labeling and model training.

[0067] S2, data optimization is performed on the lithium mineral microscopic image data after feature extraction, the optimized lithium mineral microscopic image data is divided into first part data and second part data, and the first part data is manually labeled to generate labeled sample data.

[0068] The data optimization preprocessing is performed on the lithium mineral microscopic image data, the area (number of pixels) of the connected region in the image is analyzed, the filter threshold is set to 100 pixels, and the debris region with excessively small size or excessively high separation degree in the image is removed, so that the larger region meeting the characteristics of lithium mineral is retained. This process helps to reduce the interference of irrelevant noise on subsequent analysis, thereby improving the signal-to-noise ratio of target detection, not only effectively improving the image quality and reducing unnecessary computational burden, but also helping to focus on the main target and improving the detection accuracy and robustness of the subsequent segmentation algorithm for lithium mineral. The data optimization preprocessing example is shown in Figure 2 .

[0069] The optimized lithium mineral microscopic image data is divided into first part data and second part data, for example, 20% of the optimized lithium mineral microscopic image data is selected, and artificial labeling in YOLO target detection format is performed on this part of data set, each image corresponds to a.txt file, which contains information such as target category, center coordinates of bounding box, width and height. The labeled data set will be used to train an automatic labeling model to generate a preliminary detection model weight for labeling the second part data, thereby reducing the workload of manual labeling.

[0070] S3, according to the annotated sample data, the initial version of YOLO data annotation segmentation model is trained, and the initial version of YOLO data annotation segmentation model after training is used to annotate the second part of the data.

[0071] During the training process, the initial version of YOLO data annotation segmentation model is continuously monitored, and the mAP@0.5(mean average precision) index of the model is focused on to ensure that the index reaches at least 90% or more, further ensuring the accuracy and robustness in the target detection task. After reaching this performance standard, the model obtained by training will provide reliable position information and segmentation mask for subsequent automatic annotation.

[0072] S4, based on the preset initial version of YOLO data annotation segmentation model, a preset advanced version of YOLO data annotation segmentation model is generated, the annotated sample data and the annotated second part of the data are divided into a validation set and a training set, and the preset advanced version of YOLO data annotation segmentation model is trained using the validation set and the training set to obtain the trained advanced version of YOLO data annotation segmentation model.

[0073] The preset initial version of YOLO data annotation segmentation model is improved, and the specific operation is: the channel number of the Backbone(main network) part 0-8 layers of the preset initial version of YOLO data annotation segmentation model is reduced by an equal proportion, and the reduction ratio is 1 / 2. Table 1 shows the detailed comparison of the channel number, maximum channel number and parameter amount of the initial version of YOLO data annotation segmentation model and the advanced version of YOLO data annotation segmentation model from the 0th layer to the 8th layer. This operation significantly optimizes the redundancy existing in the original baseline model; PSConv is introduced in the backbone network and neck network(Neck) part of the advanced version of YOLO data annotation segmentation model, and the Conv modules of the 0th, 1st, 3rd, 5th, 7th, 15th and 18th layers are replaced to enhance the attention to small target features; in the head network(Head) part of the advanced version of YOLO data annotation segmentation model, the detection head(Detect) is replaced by the segmentation head(Segment), and the per-channel convolution in the original segmentation head is removed to construct a lightweight Segment-Efficient segmentation head, further optimizing the calculation amount of the Head part. The advanced version of YOLO data annotation segmentation model is built through the above steps.

[0074] Table 1 Comparison of channel numbers of Backbone part before and after channel scaling

[0075]

[0076] The parameter initialization setting of model training is as follows: the batch size is set to 8, the initial learning rate is 0.01, the total number of training epochs is 150, and the patience threshold is 50, that is, if the performance on the validation set does not improve significantly within 50 rounds, the training is stopped in advance. In order to optimize the data augmentation strategy, the mosaic data augmentation will be turned off in the last 10 rounds. In terms of optimizer, the SGD optimizer (Stochastic Gradient Descent) with momentum is selected to accelerate the convergence speed of the model and avoid local minimum or saddle point problems in the training process. For the evaluation of the model, the detection frame accuracy index P box and the segmentation accuracy index P mask are used for accuracy, and the parameter quantity and the number of floating point operations (FLOPs) are selected as the quantitative standards of spatial complexity and computational complexity for lightweight.

[0077] The labeled lithium mineral microscopic image data is randomly divided into a training set and a validation set, and in some embodiments, the labeled lithium mineral microscopic image data can also be divided into a training set, a validation set and a test set. The training set is applied to the advanced YOLO data labeling and segmentation model for training. During the training process, the weight parameters of the model are continuously optimized to improve its recognition ability and robustness for different microscopic mineral textures. In order to ensure the training effect of the model, the accuracy and loss value changes on the validation set are monitored in real time, and the training strategy is adjusted in time to avoid overfitting or underfitting.

[0078] The trained advanced YOLO data labeling and segmentation model is comprehensively evaluated by the test set. By comparing with the initial version of YOLO data labeling and segmentation model, the P box and P mask of the advanced YOLO data labeling and segmentation model reach 95.0% and 97.3% respectively, which are 2.1% and 1.7% higher than the baseline model. In terms of lightweight, the parameter quantity is reduced to 1.88M and the FLOPs is reduced to 8.2G, which are 33% and 20% less than the baseline model, respectively, verifying the optimization effect of the improved modules PSConv and Segment-Efficient in feature extraction, classification accuracy and computational efficiency. See Table 2 for specific comparison.

[0079] Table 2 Comprehensive comparison of advanced model and initial model

[0080] Model \ Metric P box (%)]] P mask (%)]] Parameter amount (M) FLOPs (G) Initial version 92.9 95.6 2.81 10.2 Advanced version 95.0 97.3 1.88 8.2

[0081] S5, using the trained advanced YOLO data labeling and segmentation model to segment the lithium mineral microscopic image.

[0082] In some embodiments, a batch of dry powder samples of various components of lithium minerals (including quartz, feldspar, and lepidolite) are collected from an industrial site, and then fine photographs are taken under a 200x optical microscope to obtain high-definition microscopic images. By data optimization processing and automatic labeling of the images, a high-quality lithium mineral microscopic image dataset covering three mineral components is constructed.

[0083] In practical applications, feldspar, quartz, and lepidolite powder samples are collected from finished products after the grinding-drying process of an industrial production line, and detection samples are prepared after standardization preprocessing. The prepared powder samples are placed on the objective platform of a high-power optical microscope, and clear imaging effects are obtained through a precision focusing system. High-resolution microscopic imaging equipment is used for multi-region image acquisition, and the system records the morphological characteristics, surface texture distribution, and particle boundary information of the samples. Based on the multi-angle scanning strategy, a feature image data set covering the overall appearance of the sample is constructed to provide basic data support for subsequent algorithm training.

[0084] Next, data optimization processing is performed on the existing lithium mineral microscopic images. Small area removal and area filtering techniques are used to remove small debris by setting a filtering threshold based on the area (number of pixels) of connected regions in the image, focusing on the main ore block target.

[0085] Further, a small part of the dataset is manually annotated by using a data labeling tool. This process provides high-quality labeled data for the initial YOLO data labeling segmentation model for subsequent training.

[0086] An initial YOLO data labeling segmentation model is built, which detects targets through three parts: Backbone (main network), Neck (neck network), and Head (head network). Backbone is responsible for extracting low-level features of the image, Neck aggregates and optimizes these features to improve the detection ability of targets of different scales, and Head finally generates the detection results of the target, including position, category, and confidence. The network structure is as shown in Figure 3 .

[0087] The initial YOLO data labeling segmentation model is trained using the manually labeled part of the dataset, and the accuracy of the detection frame in the mineral component target detection is required to be at a high level, so as to provide accurate position information for the subsequent automatic labeling task to generate segmentation masks.

[0088] The target detection weight and the segmentation pre-training weight generated by training are obtained. After training in the initial version YOLO data labeling segmentation model and reaching a convergent state, the system extracts trainable parameters such as convolution kernel weight, sliding mean and variance of the batch normalization layer through a parameter serialization mechanism, and saves them as a structured storage object, finally generating a.pt format file containing the complete state of the network, laying a foundation for the subsequent automatic labeling process.

[0089] The target detection weight file and the segmentation weight file are imported, and the remaining data set is automatically labeled. The automatic labeling process includes three steps: first, the target detection weight trained in advance is used to detect the three different component minerals in the data set, and a rectangular detection box is formed according to the generated coordinates, and the corresponding category is labeled; then, the segmentation weight is used to generate a segmentation mask with a category label for the ore block in the detection box; finally, the generated labeling result is saved as a text file (.txt format), and the data is divided according to a certain scientific proportion for subsequent use.

[0090] Optionally, as shown in Figure 4 the preset initial version YOLO data labeling segmentation model, a preset advanced version YOLO data labeling segmentation model is generated, including:

[0091] S401, the channel number of the backbone network part of the preset initial version YOLO data labeling segmentation model is scaled by equal proportion, and the channel number of each backbone layer is reduced by half.

[0092] S402, all conventional convolutions in the preset initial version YOLO data labeling segmentation model backbone network and neck network are replaced with lightweight windmill-shaped convolutions.

[0093] S403, all channel-by-channel convolutions in the head network of the preset initial version YOLO data labeling segmentation model are removed to construct a lightweight Segment-Efficient segmentation head.

[0094] The advanced version YOLO data labeling segmentation model is built, and its structure is as shown in Figure 5 . The model aims to better capture the texture details of the ore block, enhance the extraction ability of different scale features in the image, effectively distinguish the differences between different types of ore blocks in the lithium mineral microscopic image, and improve the segmentation accuracy, while ensuring that the model is lightweight and real-time. The present application makes the following key improvements for the lithium mineral microscopic image segmentation task:

[0095] First, the channel number of the Backbone part of the initial version of the YOLO data annotation segmentation model is scaled proportionally while keeping the global depth, width and scale unchanged. Specifically, the maximum channel number is limited to 512, and the channel number of each backbone layer (0th to 8th layer) is reduced by half. This makes the advanced version of the YOLO data annotation segmentation model more lightweight and compact than the original baseline model of the initial version of the YOLO data annotation segmentation model.

[0096] Secondly, a lightweight pinwheel-shaped convolution PSConv is used to replace all regular convolutions in the entire network as the improved network downsampling operation. The workflow diagram of PSConv is shown in Figure 6 . Specifically, let the input tensor be where c1, h1, w1 represent the channel size, height and width respectively, and to enhance training stability and speed, batch normalization (BN) and sigmoid linear unit (SiLU) are applied after each convolution. PSConv first performs four-way parallel convolution, and the calculation formula is as follows:

[0097]

[0098] where, is the convolution operator, i = 1, 2, 3, 4 respectively represent the four directions, is the convolution kernel with size k x d, and the output channel is c ′ . The padding parameter P represents the number of padding pixels in the left, right, top and bottom directions. The channel number c ′ , height h ′ and width w ′ of the output feature map after the first layer of four-way staggered convolution are related to the input feature map as follows: h ′ = 1 + h1 / s, w ′ = 1 + w1 / s, c ′ = c2 / 4. Where c2 is the channel number of the final output feature map of PSConv, and s is the convolution step. The result of the first layer of convolution is concatenated (Concat) to output the calculation formula as follows:

[0099]

[0100] Finally, the concatenated tensor is normalized by the convolution kernel W(c2, 2, 2) without padding. The height and width of the output feature map are adjusted to the preset values h2 and w2, where h2 = h1 / s, w2 = w1 / s. Its role is to analyze the contribution of different convolution directions through the channel attention mechanism. The final output Y(c2, h2, w2) is calculated as follows:

[0101]

[0102] PSConv utilizes group convolution, which can only increase a small amount of parameters while significantly enhancing large receptive fields. By asymmetric padding, it creates more symmetric horizontal and vertical non-block convolutions, which expand the receptive field and better improve the features of small targets, thus enhancing the ability of layer feature extraction and improving the performance of detection and segmentation.

[0103] Finally, the initial version of YOLO data annotation segmentation model is removed from the head part of the channel-by-channel convolution (Depthwise Convolution), and a lightweight Segment-Efficient segmentation head is constructed. The head structure of the initial version of YOLO data annotation segmentation model is as shown in Figure 7 , and the Segment-Efficient structure is as shown in Figure 8 . Figure 8 The improvement optimizes the segmentation calculation process of the head, reduces the parameter quantity and calculation complexity, and through the simplified structure and more efficient convolution layer, it can significantly improve the inference speed and the practicability of the model under the premise of ensuring the detection accuracy. The network structure design of the advanced version of YOLO data annotation segmentation model fully utilizes the advantages of the initial version of YOLO data annotation segmentation model and the improved module in terms of real-time performance, calculation efficiency and detection accuracy, so that it has strong applicability in the segmentation task of lithium mineral microscopic images, especially suitable for efficient and accurate automatic segmentation and grade evaluation of ores on edge devices with limited computing resources.

[0104] Since the channel number of the Backbone part is scaled by the same proportion, the advanced version of YOLO data annotation segmentation model is more simplified than the initial version of YOLO data annotation segmentation model, which significantly reduces the calculation amount and parameter quantity. Secondly, the Backbone part introduces PSConv (Pinwheel-shaped Convolution), which creates horizontal and vertical convolution kernels through asymmetric padding and pinwheel structure, thereby expanding the receptive field without significantly increasing the parameters, especially suitable for capturing the context information of small targets. In addition, PSConv optimizes the parameter efficiency while improving the performance, which is suitable for resource-limited environments (such as embedded devices). Finally, the segmentation head is optimized in the Head part, and the Segment-Efficient module is constructed, which improves the generation of detection boxes and segmentation masks, further reducing the calculation amount. Through the above methods, the improved model realizes significant optimization in feature extraction and lightweight design, improves the accuracy and efficiency of lithium mineral segmentation, and is suitable for real-time applications in industrial sites.

[0105] Optionally, as Figure 9As shown, the use of the training set and the training set is used to train the preset advanced version YOLO data annotation segmentation model, including:

[0106] S901, each parameter of the preset advanced version YOLO data annotation segmentation model is initialized, wherein the parameters include the total number of training rounds, the number of samples processed per round, the patience threshold and the initial learning rate.

[0107] S902, using the training set to train the preset advanced version YOLO data annotation segmentation model, after each training cycle, using the validation set to monitor the performance of the advanced version YOLO data annotation segmentation model, in the case of continuous improvement of the accuracy of the advanced version YOLO data annotation segmentation model and gradual decrease of the loss value, continue to train, in the case of decrease of the accuracy of the advanced version YOLO data annotation segmentation model or increase of the loss value, terminate the training.

[0108] After the training is completed, the trained advanced version YOLO data annotation segmentation model can be used to evaluate the performance of the test set.

[0109] Each configuration parameter of the advanced version YOLO data annotation segmentation model is initialized, including determining the total number of training rounds (epoch), the number of samples processed per batch (batch size), setting the patience threshold (patience) and the initial learning rate (learning rate), selecting the appropriate gradient descent-based optimization algorithm and configuring the related hyperparameters. In addition, it is also necessary to define quantitative indicators for comprehensive evaluation of model detection accuracy, segmentation accuracy and efficiency to ensure real-time monitoring and adjustment during the training process, thereby further optimizing the overall performance of the model.

[0110] The training set is used to train the advanced version YOLO data annotation segmentation model, and the model performance is monitored in real time through the validation set. After each training cycle, the performance of the model on the validation set is immediately evaluated, focusing on the change of the accuracy and the trend of the loss value. If the accuracy continues to improve and the loss value gradually decreases, the training will continue; if the reverse trend of the decrease of the accuracy or the increase of the loss value is found, the training should be terminated in time to prevent overfitting and save computing resources.

[0111] The trained advanced version YOLO data annotation segmentation model is obtained, and the performance is evaluated on the pre-divided test set. By comparing with the initial version YOLO data annotation segmentation model, the differences in various experimental indicators are analyzed, the effects of the improved modules (such as PSConv and Segment-Efficient) are summarized, and the feasibility in practical application is discussed.

[0112] The final trained segmentation model is deployed on the lithium mineral microscopic image segmentation intelligent equipment, so that it can analyze and process the quality information of lithium minerals in real time.

[0113] Optionally, the feature extraction on the lithium mineral microscopic image data comprises:

[0114] The morphological features, surface texture distribution and particle boundary information of the lithium mineral microscopic image data are extracted.

[0115] Optionally, the data optimization on the lithium mineral microscopic image data comprises:

[0116] The small area removal and area filtering techniques are used to set the filtering threshold by analyzing the area of the connected region in the lithium mineral microscopic image data, and to remove the over-separated small debris in the lithium mineral microscopic image data.

[0117] Optionally, the preset initial version YOLO data labeling segmentation model has a backbone network responsible for extracting low-level features of the lithium mineral microscopic image data. The neck network of the preset initial version YOLO data labeling segmentation model is responsible for aggregating and optimizing the low-level features of the lithium mineral microscopic image data to improve the detection ability of different scale targets. The head network of the preset initial version YOLO data labeling segmentation model is responsible for generating the detection result of the target.

[0118] Optionally, as shown in Figure 10 the second part of data using the trained initial version YOLO data labeling segmentation model comprises:

[0119] S1001, according to the trained target detection weight, using the initial version YOLO data labeling segmentation model to detect the target of the second part of data, and obtaining the target coordinate.

[0120] S1002, generating a detection frame according to the target coordinate.

[0121] S1003, according to the trained segmentation weight, using the initial version YOLO data labeling segmentation model to generate a segmentation mask with a class label for the ore block image in the detection frame.

[0122] According to the advanced version YOLO data labeling segmentation model data, the segmentation mask precision and confidence curve is drawn as shown in Figure 11 the segmentation effect is shown in Figure 12 It is shown that the advanced version YOLO data labeling segmentation model has high accuracy and real-time detection ability, and fully meets the deployment requirements of edge embedded devices.

[0123] As shown in Figure 13As shown, the present application also provides a lithium mineral microscopic image segmentation device for implementing the lithium mineral microscopic image segmentation method provided by the embodiments of the present application, and the device comprises:

[0124] The acquisition module 1301 is configured to acquire lithium mineral microscopic image data and perform feature extraction on the lithium mineral microscopic image data.

[0125] The labeling module 1302 is configured to perform data optimization on the lithium mineral microscopic image data after feature extraction, divide the optimized lithium mineral microscopic image data into first part data and second part data, and manually label the first part data to generate labeled sample data.

[0126] The initial version model training module 1303 is configured to train a preset initial version YOLO data labeling and segmentation model according to the labeled sample data, and label the second part data using the trained initial version YOLO data labeling and segmentation model.

[0127] The advanced version model training module 1304 is configured to generate a preset advanced version YOLO data labeling and segmentation model based on the preset initial version YOLO data labeling and segmentation model, divide the sample data and the labeled second part data into a verification set and a training set, train the preset advanced version YOLO data labeling and segmentation model using the verification set and the training set, and obtain a trained advanced version YOLO data labeling and segmentation model.

[0128] The segmentation module 1305 is configured to segment the lithium mineral microscopic image using the trained advanced version YOLO data labeling and segmentation model.

[0129] In another aspect, a lithium mineral microscopic image segmentation device is provided, and the lithium mineral microscopic image segmentation device comprises:

[0130] A processor.

[0131] A memory having computer readable instructions stored thereon, wherein the computer readable instructions, when executed by the processor, implement the method provided by the embodiments of the present application.

[0132] In another aspect, a computer readable storage medium is provided, and the computer readable storage medium stores program codes, wherein the program codes can be called by a processor to execute the method provided by the embodiments of the present application.

[0133] The technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0134] The embodiment of the present application can realize the following beneficial effects by segmenting the lithium mineral microscopic image through the advanced YOLO data labeling segmentation model: (1) automatic labeling method: a small amount of data labeled by hand is used to train the detection model, and then the obtained target detection and segmentation pre-training weight is used to automatically label a large-scale data set, thereby effectively reducing the workload of a large number of instance labeling in the image segmentation task. (2) balance between precision and efficiency: by using the advanced YOLO data labeling segmentation model, the calculation efficiency of the model is balanced while ensuring the detection accuracy and segmentation accuracy. The model can achieve high precision in practical application, and also has sufficient real-time processing capability, which adapts to the application requirements in a dynamic environment. (3) industrial application deployment: the advanced YOLO data labeling segmentation model of the present application can be deployed in the lithium mica ore flotation effect evaluation intelligent equipment, and supports real-time industrial field analysis and detection. It can greatly improve the recognition efficiency of effective components of lithium minerals, and provides intelligent support for mining production and resource optimization. (4) mineral texture focusing: by introducing the windmill-shaped convolution, the ability of the model to extract and identify the texture details of minerals in complex microscopic images is significantly enhanced, especially in dealing with subtle textures and complex backgrounds, which can effectively improve the segmentation accuracy; (5) efficient model optimization: by optimizing the redundancy of channels in the network backbone and the redundancy of head operations, unnecessary calculation burden is reduced, and the parameter quantity and calculation complexity of the model are reduced. This improvement ensures that the model can maintain high efficient calculation performance in real-time detection tasks, while reducing the demand for hardware resources.

[0135] Figure 14 is a structural schematic diagram of a lithium mineral microscopic image segmentation device provided by the embodiment of the present application, as Figure 14 indicated, the lithium mineral microscopic image segmentation device 1410 can include a first processor 2001.

[0136] Optionally, the lithium mineral microscopic image segmentation device 1410 can further include a memory 2002 and a transceiver 2003.

[0137] The first processor 2001, the memory 2002 and the transceiver 2003 can be connected through a communication bus.

[0138] The various constituent components of the lithium mineral microscopic image segmentation device 1410 will be specifically introduced as follows: Figure 14

[0139] ​The first processor 2001 is the control center of the lithium mineral microscopic image segmentation device 1410, and can be one processor or a collective term of multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), and can also be an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present application, such as one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs).

[0140] Optionally, the first processor 2001 can execute various functions of the lithium mineral microscopic image segmentation device 1410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0141] In a specific implementation, as an embodiment, the first processor 2001 can include one or more CPUs, such as the CPU0 and the CPU1 shown in FIG. 1. Figure 14

[0142] In a specific implementation, as an embodiment, the lithium mineral microscopic image segmentation device 1410 can also include multiple processors, such as the first processor 2001 and the second processor 2004 shown in FIG. 1. Each of these processors can be a single-CPU or a multi-CPU. The processor here can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions). Figure 14

[0143] The memory 2002 is configured to store software programs for implementing the schemes of the present application, and the first processor 2001 is configured to control the execution of the software programs. The specific implementation can refer to the above-mentioned method embodiments, and will not be repeated here.

[0144] ​​Alternatively, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently and accessed through the interface circuit ( Figure 14 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0145] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0146] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 14 (not shown separately in the figure). The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0147] Optionally, the transceiver 2003 may be integrated with the first processor 2001 or may exist independently and be connected to the lithium mineral microscopic image segmentation device 1410 through the interface circuit ( Figure 14 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0148] It should be noted that Figure 14 The structure of the lithium mineral microscopic image segmentation device 1410 shown in the figure does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0149] In addition, the technical effects of the lithium mineral microscopic image segmentation device 1410 can refer to the technical effects of the multimodal emotion recognition method described in the above method embodiment, and will not be repeated here.

[0150] It is to be understood that the first processor 2001 in the embodiments of the present application can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or can also be any conventional processor.

[0151] It is also to be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically EPROM (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM) and direct rambus RAM (DR RAM).

[0152] The above-described embodiments can be implemented in whole or in part by software, hardware (e.g., circuitry), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, electric drive or data center to another website, computer, electric drive or data center by infrared, microwave, etc. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as an electric drive, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0153] It should be understood that the term "and / or" herein merely describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but can also represent an "and / or" relationship, which can be understood in the context before and after.

[0154] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0155] It should be understood that in various embodiments of the present application, the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0156] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0157] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0158] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0159] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0160] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0161] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, an electric driver, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0162] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A lithium mineral microscopic image segmentation method, characterized in that: include: Acquire lithium mineral microscopic image data and perform feature extraction on the lithium mineral microscopic image data; Performing data optimization on the lithium mineral microscopic image data after feature extraction, dividing the optimized lithium mineral microscopic image data into a first part of data and a second part of data, and manually annotating the first part of data to generate annotated sample data; Based on the labeled sample data, the preset initial version of the YOLO data labeling and segmentation model is trained, and the trained initial version of the YOLO data labeling and segmentation model is used to label the second part of the data; Based on the preset initial version of the YOLO data annotation and segmentation model, a preset advanced version of the YOLO data annotation and segmentation model is generated, the labeled sample data and the second part of the labeled data are divided into a validation set and a training set, and the preset advanced version of the YOLO data annotation and segmentation model is trained using the validation set and the training set to obtain the trained advanced version of the YOLO data annotation and segmentation model; Use the trained advanced version of YOLO data annotation and segmentation model to perform lithium mineral microscopic image segmentation.

2. The method according to claim 1, characterized in that The method of generating a preset advanced version of the YOLO data annotation segmentation model based on the preset initial version of the YOLO data annotation segmentation model includes: The number of channels in the backbone network of the preset initial version of the YOLO data annotation and segmentation model is scaled proportionally, and the number of channels in each backbone layer is reduced by half; Use lightweight windmill convolution to replace all conventional convolutions in the backbone network and neck network of the preset initial version of the YOLO data annotation and segmentation model; All channel-by-channel convolutions in the head network of the preset initial version of the YOLO data annotation and segmentation model are removed to build a lightweight Segment-Efficient segmentation head.

3. The method according to claim 1, characterized in that The use of the validation set and training set to train the preset advanced version of the YOLO data annotation segmentation model includes: Initialize various parameters of the preset advanced version of YOLO data annotation and segmentation model, including the total number of training rounds, the number of samples processed in each round, the patience threshold, and the initial learning rate; The preset advanced version of the YOLO data annotation segmentation model is trained using the training set. After each training cycle, the performance of the advanced version of the YOLO data annotation segmentation model is monitored using the validation set. If the accuracy of the advanced version of the YOLO data annotation segmentation model continues to improve and the loss value gradually decreases, the training continues. If the accuracy of the advanced version of the YOLO data annotation segmentation model decreases or the loss value increases, the training is terminated.

4. The method according to claim 1, wherein The feature extraction of lithium mineral microscopic image data includes: Extract morphological characteristics, surface texture distribution and particle boundary information from lithium mineral microscopic image data.

5. The method according to claim 1, wherein The data optimization of lithium mineral microscopic image data includes: Small area removal and area filtering techniques are used to set the filtering threshold by analyzing the area of ​​the connected regions in the lithium mineral microscopic image data, thereby removing over-separated small debris in the lithium mineral microscopic image data.

6. The method according to claim 1, characterized in that The backbone network of the preset initial version of the YOLO data annotation and segmentation model is responsible for extracting low-level features of lithium mineral microscopic image data; The neck network of the preset initial version of the YOLO data annotation and segmentation model is responsible for aggregating and optimizing low-level features of lithium mineral microscopic image data to improve the detection capability of targets of different scales; The head network of the preset initial version of the YOLO data annotation and segmentation model is responsible for generating the detection results of the target.

7. The method according to claim 1, characterized in that The second part of the data is labeled using the trained initial version of the YOLO data labeling and segmentation model, including: According to the trained target detection weights, the initial version of the YOLO data annotation and segmentation model is used to perform target detection on the second part of the data to obtain the target coordinates; Generate a detection frame based on the target coordinates; Based on the trained segmentation weights, the initial version of the YOLO data annotation segmentation model is used to generate a segmentation mask with category labels for the mineral block image within the detection box.

8. A lithium mineral microscopic image segmentation device, the lithium mineral microscopic image segmentation device is used to implement the lithium mineral microscopic image segmentation method according to any one of claims 1 to 7, characterized in that: The device comprises: An acquisition module is used to acquire lithium mineral microscopic image data and perform feature extraction on the lithium mineral microscopic image data; a labeling module, configured to optimize the lithium mineral microscopic image data after feature extraction, divide the optimized lithium mineral microscopic image data into a first portion of data and a second portion of data, and manually label the first portion of data to generate labeled sample data; The initial model training module is used to train the preset initial version YOLO data annotation and segmentation model based on the labeled sample data, and use the trained initial version YOLO data annotation and segmentation model to annotate the second part of the data; The advanced model training module is used to generate a preset advanced version of the YOLO data annotation and segmentation model based on the preset initial version of the YOLO data annotation and segmentation model, divide the sample data and the second part of the annotated data into a validation set and a training set, and use the validation set and the training set to train the preset advanced version of the YOLO data annotation and segmentation model to obtain a trained advanced version of the YOLO data annotation and segmentation model; A segmentation module is used to segment lithium mineral microscopic images using the trained advanced YOLO data annotation segmentation model.

9. A lithium mineral microscopic image segmentation device, characterized in that: The lithium mineral microscopic image segmentation device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images