Image processing method and system for mine lamp, storage medium and mine lamp
Through the depth generation model and multi-scale convolutional network, the light correction and reflection enhancement of the images collected by the mine lamp are solved, and the image quality of traditional mine lamps in low-light environments is achieved, clearer and even image display is achieved, and the mine is improved The effectiveness of monitoring and security management.
Patent Information
- Application Number
- CN202411279646.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional mineral lamps cannot ensure sufficient brightness and light uniformity in low-light, dust and complex reflection environments, resulting in limited view of miners and poor image quality, which affects the monitoring effect and safety management inside the mine.
Using image processing method, the lighting components are corrected through the depth generation model, and the reflected components are enhanced by using a multi-scale convolutional network and a deep denoising autoencoder, separated and fused the lighting and reflection components to generate a clear and uniform target image. .
It effectively improves image brightness and detail clarity in low-light environments, reduces light unevenness, removes noise and artifacts, improves the imaging quality of the mine lamp camera, and ensures clear, accurate monitoring and safe management of the internal environment of the mine.
Smart Images

Figure CN118941455A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of lighting equipment, and in particular to an image processing method, system, storage medium and mining lamp for mining lamps. Background Art
[0002] In the mine working environment, the lighting conditions are usually complex and changeable. Low light, dust and strong light reflections greatly limit the field of vision in the working environment. Traditional mining lamp lighting systems usually use a simple brightness adjustment method, which is difficult to automatically adjust the brightness according to the dynamic lighting conditions in the mine, resulting in limited vision of miners, unable to clearly observe the working area and its surrounding environment, thus affecting work efficiency and safety.
[0003] In low-light conditions, miners often rely on the lighting provided by mining lamps, but traditional mining lamps cannot guarantee sufficient brightness and uniformity of light in environments with high dust and complex reflective surfaces. In this case, miners will see obvious dark areas or over-reflected bright spots in their field of vision, affecting their normal work judgment. More importantly, the images captured by the cameras equipped in the mining lamps often have problems such as insufficient brightness, low contrast, and uneven lighting, making it difficult to clearly present the situation inside the mine, which seriously affects the monitoring effect and safety management inside the mine. Summary of the invention
[0004] Based on this, the purpose of the present invention is to provide an image processing method, system, storage medium and mining lamp for a mining lamp, so as to fundamentally solve the problem of poor image quality in the existing low-light environment.
[0005] An image processing method for a mining lamp according to an embodiment of the present invention includes:
[0006] Collect image data of the current environment, and perform image decomposition on the collected image data to obtain illumination component and reflection component;
[0007] Performing illumination correction on the illumination component according to the deep generative model to generate a corrected illumination component;
[0008] Performing reflection enhancement on the reflection component according to a multi-scale convolutional network and a deep denoising autoencoder to generate an enhanced reflection component;
[0009] The corrected illumination component and the enhanced reflection component are fused to obtain the target image data.
[0010] In addition, the image processing method for a mining lamp according to the above embodiment of the present invention may also have the following additional technical features:
[0011] Furthermore, the step of performing image decomposition on the collected image data to obtain illumination components and reflection components includes:
[0012] The collected current image data is converted into a grayscale image, and the grayscale image is logarithmically transformed to obtain a logarithmic domain image;
[0013] Gaussian filtering is performed on the logarithmic domain image after logarithmic transformation to obtain the logarithmic value of the estimated illumination component, and the estimated illumination component is converted from the logarithmic domain back to the linear domain;
[0014] The reflection component is calculated based on the grayscale value of the current image data and the estimated illumination component.
[0015] Furthermore, the step of performing illumination correction on the illumination component according to the deep generative model to generate the corrected illumination component includes:
[0016] Constructing a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component;
[0017] Define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator;
[0018] Train a deep generative model. In each training iteration, the generator generates new illumination components, the discriminator evaluates the authenticity of the generated illumination components, calculates the losses of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer.
[0019] Input the illumination component into the trained generator to generate a corrected illumination component;
[0020] Applies contrast and brightness adjustments to the resulting lighting components.
[0021] Furthermore, the step of performing reflection enhancement on the reflection component according to the multi-scale convolutional network and the deep denoising autoencoder to generate the enhanced reflection component includes:
[0022] Extract features of different scales in the reflection component according to the multi-scale convolutional network, fuse the features extracted from different scales to obtain a comprehensive feature map, and convert the fused comprehensive feature map into an enhanced reflection component;
[0023] According to the encoder in the deep denoising autoencoder, the reflected component enhanced by the multi-scale convolutional network is compressed into a low-dimensional feature representation and denoised, and according to the decoder in the deep denoising autoencoder, the enhanced reflected component is restored from the low-dimensional feature representation.
[0024] Furthermore, the step of fusing the corrected illumination component and the enhanced reflection component to obtain target image data includes:
[0025] Normalizing the corrected illumination component and the enhanced reflection component;
[0026] Multiply the normalized illumination component and reflection component pixel by pixel to generate target image data;
[0027] Perform contrast adjustment and color correction on the target image data.
[0028] Another embodiment of the present invention is to provide an image processing system for a mining lamp, the system comprising:
[0029] A data processing module is used to collect image data of the current environment and perform image decomposition on the collected image data to obtain illumination components and reflection components;
[0030] An illumination correction module, used for performing illumination correction on the illumination component according to the deep generation model to generate a corrected illumination component;
[0031] A reflection enhancement module, used to enhance the reflection component according to a multi-scale convolutional network and a deep denoising autoencoder to generate an enhanced reflection component;
[0032] The data fusion module is used to fuse the corrected illumination component and the enhanced reflection component to obtain the target image data.
[0033] Furthermore, the data processing module includes:
[0034] A data conversion unit, used for converting the collected current image data into a grayscale image, and performing logarithmic transformation on the grayscale image to obtain a logarithmic domain image;
[0035] an illumination component calculation unit, configured to perform Gaussian filtering on the logarithmic domain image after logarithmic transformation to obtain a logarithmic value of the estimated illumination component, and to convert the estimated illumination component from the logarithmic domain back to the linear domain;
[0036] The reflection component calculation unit is used to calculate the reflection component according to the gray value of the current image data and the estimated illumination component.
[0037] Furthermore, the illumination correction module includes:
[0038] A module building unit for building a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component;
[0039] The data definition unit is used to define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator;
[0040] The model training unit is used to train the deep generative model. In each training iteration, the generator generates a new illumination component, the discriminator evaluates the authenticity of the generated illumination component, calculates the loss of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer;
[0041] An illumination component generating unit, used for inputting the illumination component into the trained generator to generate a corrected illumination component;
[0042] The illumination component adjustment unit is used to adjust the contrast and brightness of the generated illumination components.
[0043] Another embodiment of the present invention aims to provide a storage medium storing a program, which, when executed by a processor, implements the image processing method for a mining lamp as described above.
[0044] Another embodiment of the present invention aims to provide a mining lamp, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor implements the image processing method for the mining lamp as described above when executing the program.
[0045] The image processing method for a mining lamp provided by an embodiment of the present invention can effectively enhance the image brightness and detail clarity in a low-light environment by separately processing the illumination component and the reflection component, improve the problems of uneven illumination and insufficient brightness, and make the key details in the mine environment more clearly visible; by correcting the illumination component using a deep generative model, the illumination effect can be dynamically adjusted to reduce the uneven illumination phenomenon in the collected image caused by the complex environment of the mine; the reflection component is enhanced by a multi-scale convolutional network and a deep denoising autoencoder, and the noise and artifacts caused by interference such as light reflection and mine dust are effectively removed, so that the final generated image is cleaner, richer in details, and higher in image contrast; through the enhanced image data, the imaging quality of the camera in the mining lamp can be improved, so that the monitoring system can more clearly and accurately reflect the internal environment of the mine and the mining operation conditions, effectively ensure the safety management and operation monitoring of the mine, and solve the problem of poor image quality in the existing low-light environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 is a flow chart of an image processing method for a mining lamp in a first embodiment of the present invention;
[0047] Figure 2 is a schematic structural diagram of an image processing system for a mining lamp in a second embodiment of the present invention;
[0048] Figure 3 It is a schematic structural diagram of a mining lamp in a third embodiment of the present invention;
[0049] The following specific implementations will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION
[0050] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.
[0051] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.
[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0053] Embodiment 1
[0054] See also Figure 1 , which shows an image processing method for a miner's lamp in a first embodiment of the present invention. For the sake of convenience, only the part related to the embodiment of the present invention is shown. The image processing method for a miner's lamp provided by the embodiment of the present invention includes:
[0055] Step S10, collecting image data of the current environment, and performing image decomposition on the collected image data to obtain an illumination component and a reflection component;
[0056] Among them, in one embodiment of the present invention, first, it is necessary to use a camera or other imaging device to capture an image of the current environment. This is usually achieved by calling the API of the image acquisition hardware. A common method is to use a camera interface, such as OpenCV, to capture images in real time. Further, the captured image data is decomposed to obtain an illumination component and a reflection component, which usually depends on an image decomposition algorithm, especially a decomposition method based on the Retinex theory. Among them, the Retinex theory proposes that the brightness of an image is determined by two main factors: Reflectance Component: represents the inherent reflection property or color of the surface of an object. Illumination Component: represents the intensity of ambient light. According to the Retinex theory, the brightness of an image I(x,y) is the product of the reflection component R(x,y) and the illumination component L(x,y), that is: I(x,y) = R(x,y)·L(x,y), where I(x,y) is the pixel intensity at the coordinate (x,y). R(x,y) is the reflection component at the coordinate (x,y), and L(x,y) is the illumination component at the coordinate (x,y). At this time, the above-mentioned step of performing image decomposition on the collected image data to obtain the illumination component and the reflection component includes:
[0057] The collected current image data is converted into a grayscale image, and the grayscale image is logarithmically transformed to obtain a logarithmic domain image;
[0058] Gaussian filtering is performed on the logarithmic domain image after logarithmic transformation to obtain the logarithmic value of the estimated illumination component, and the estimated illumination component is converted from the logarithmic domain back to the linear domain;
[0059] The reflection component is calculated based on the grayscale value of the current image data and the estimated illumination component.
[0060] Specifically, the color image is converted into a grayscale image. For color images, the weighted average method is usually used to convert the RGB image into a grayscale image. For example: L = 0.299R + 0.587G + 0.114B, where R, G, and B are the values of the red, green, and blue channels of each pixel in the image. The grayscale image is then logarithmically transformed to convert the multiplication relationship into an addition relationship. This helps with subsequent decomposition processing. Use the logarithmic transformation formula: log(I(x,y)) = log(R(x,y)) + log(L(x,y)). Gaussian filtering is then applied to the logarithmic transformed image to estimate the illumination component. Gaussian filtering can smooth the image, remove local details, and retain global illumination information. The formula is as follows:
[0061] log(L(x,y))=G(x,y)*log(I(x,y))
[0062] Where G(x,y) is the Gaussian kernel and * indicates the convolution operation. The result log(L(x,y)) is the logarithm of the estimated illumination component. The estimated illumination component is then converted from the logarithmic domain back to the linear domain. Specifically, an exponential function is used to restore it from the logarithmic domain back to the linear domain:
[0063] L(x,y)=exp(log(L(x,y)))
[0064] The reflection component is then calculated using the current image data and the estimated illumination component. The formula is as follows:
[0065]
[0066] where ∈ is a small constant (e.g. 1×10 -8 ) to avoid division by zero errors.
[0067] Furthermore, in addition to the above-mentioned Retinex theory decomposition method to decompose the collected current image data to obtain the reflection component and the illumination component, a deep learning method (such as a Retinex-based neural network) can also be used for image decomposition. Specifically, first use an image data set containing real reflection and illumination components to train a deep neural network module. Then the collected image is input into the trained deep neural network model, and the deep neural network module outputs the corresponding decomposition to obtain the reflection component and illumination component. Commonly used network structures for denoising include the Retinex-Net network structure and the Deep Retinex network structure, in which Retinex-Net is based on a deep convolutional neural network (CNN), and extracts the reflection component and illumination component from the input image through network learning. Deep Retinex is a Retinex model based on deep learning, which combines multi-scale feature extraction and generative adversarial network (GAN) for image decomposition.
[0068] Step S20, performing illumination correction on the illumination component according to the depth generation model to generate a corrected illumination component;
[0069] In one embodiment of the present invention, the step of performing illumination correction on the illumination component according to the depth generation model to generate the corrected illumination component includes:
[0070] Constructing a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component;
[0071] Define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator;
[0072] Train a deep generative model. In each training iteration, the generator generates new illumination components, the discriminator evaluates the authenticity of the generated illumination components, calculates the losses of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer.
[0073] Input the illumination component into the trained generator to generate a corrected illumination component;
[0074] Applies contrast and brightness adjustments to the resulting lighting components.
[0075] Specifically, a deep generative model is first selected, among which commonly used deep generative models include generative adversarial networks (GANs) and variational autoencoders (VAEs). In the illumination correction task, GANs are particularly suitable for generating corrected illumination components that match the target illumination components, so in the embodiment of the present invention, generative adversarial networks (GANs) are selected as deep generative models. The generative adversarial network (GANs) includes a generator and a discriminator, wherein the generator network is responsible for generating the corrected illumination component, and the discriminator network is responsible for distinguishing the difference between the generated illumination component and the real illumination component, ensuring that the image generated by the generator is as realistic as possible.
[0076] Specifically, the generator network is composed of a convolution layer, an upsampling layer, and an activation function to generate a corrected illumination component. Its input is the illumination component, which is processed by multiple convolution layers, activation functions (such as ReLU) and upsampling layers, so that the corrected illumination component is finally generated and output. The discriminator network is composed of a convolution layer and a fully connected layer to evaluate the authenticity of the generated illumination component. Its input is the corrected illumination component or the real illumination component, which is classified by multiple convolution layers, activation functions (such as LeakyReLU) and fully connected layers, so that the probability of whether the generated illumination component is a real image is finally output. Further, a loss function and an optimizer are defined, wherein the loss function includes a generator loss function and a discriminator loss function, wherein the generator loss function is that the generator hopes that the generated illumination component is as close to the real illumination component as possible, thereby deceiving the discriminator, making it think that the generated image is real, and making it difficult for the discriminator to distinguish between true and false. The discriminator loss function is that the discriminator hopes to accurately distinguish the generated illumination component from the real illumination component, ensuring that the generated illumination component is more realistic visually. At this time, the generator loss function and the discriminator loss function are used to guide the training process. The optimizer is used to update the weights of the generator and the discriminator. Further, to train the deep generative model, first, the generator and the discriminator are initialized using randomly initialized or pre-trained weights. The network parameters are optimized by alternately training the generator and the discriminator. The generator generates illumination components, the discriminator evaluates its authenticity, and feeds back to the generator to adjust the generation results. The optimizer is used to update the weights of the generator and the discriminator to minimize the loss function. In each training iteration, the generator is used to generate new illumination components, the discriminator is used to evaluate the authenticity of the generated illumination components, the losses of the generator and the discriminator are calculated, and the weights of the generator and the discriminator are updated by the optimizer. Further, the illumination components to be corrected are input into the trained generator to generate the corrected illumination components. Finally, the contrast and brightness of the generated illumination components are adjusted to optimize the final visual effect.
[0077] Among them, because deep generative models (such as generative adversarial networks GANs or variational autoencoders VAEs) can learn high-level features of images through multiple convolutional layers and nonlinear activation functions, the model can understand and generate more complex image structures and details. At this time, the deep generative model can capture the contextual information in the image and generate more detailed and realistic illumination components. It can adaptively adjust the illumination components, which helps to improve the clarity and quality of the image. At the same time, when the generator network processes the image, it can restore the details of the image through operations such as deconvolution (upsampling), which may be lost or blurred under low-light conditions. At the same time, the deep generative model can adjust the contrast of the illumination component, making the difference between brightness and darkness in the image more obvious, thereby improving the visual effect. Therefore, the deep generative model can adaptively adjust the illumination component, improve the brightness and contrast of the image, and make the details in the image more clearly visible.
[0078] At the same time, the deep generative model can learn the lighting patterns under different lighting conditions and make adaptive adjustments to uneven lighting. By comparing the real and generated lighting components, the deep generative model can identify and correct the inconsistencies in the lighting distribution. During the training process, the generator will learn how to balance the lighting in different areas so that the generated lighting components are more uniform and consistent overall. The discriminator in adversarial training forces the generator to generate more realistic lighting components by continuously feeding back the generator's generation results. Its adversarial training method makes the generated lighting components more consistent and natural overall. Therefore, it can effectively improve the unevenness of lighting in the image and make the lighting distribution of the entire image more balanced.
[0079] At the same time, during the training process of the deep generative model, by learning high-quality image data, the model can identify and remove noise generated under low-light conditions. The generator can reduce the noise in the image by optimizing the output in a high-dimensional space. The generator usually has a strong smoothing ability, which can remove small noise and artifacts in the image, making the image look cleaner. At the same time, the deep generative model can identify artifacts generated in low-light images and repair them. By learning the lighting and structural patterns in real images, the generator can effectively remove artifacts and generate more realistic images. At the same time, the loss function used by the deep generative model can guide the model to generate high-quality, low-noise images. Therefore, the noise and artifacts generated by low-light conditions can be removed, and a cleaner and more realistic lighting component can be generated. At the same time, the corrected lighting component can significantly improve the visual effect of the image, making the image look more natural and realistic.
[0080] Step S30, performing reflection enhancement on the reflection component according to the multi-scale convolutional network and the deep denoising autoencoder to generate an enhanced reflection component;
[0081] Among them, in one embodiment of the present invention, the step of performing reflection enhancement on the reflection component according to the multi-scale convolutional network and the deep denoising autoencoder to generate the enhanced reflection component includes:
[0082] Extract features of different scales in the reflection component according to the multi-scale convolutional network, fuse the features extracted from different scales to obtain a comprehensive feature map, and convert the fused comprehensive feature map into an enhanced reflection component;
[0083] According to the encoder in the deep denoising autoencoder, the reflected component enhanced by the multi-scale convolutional network is compressed into a low-dimensional feature representation and denoised, and according to the decoder in the deep denoising autoencoder, the enhanced reflected component is restored from the low-dimensional feature representation.
[0084] Specifically, a multi-scale convolutional network is used to extract features of different scales from the reflection component to enhance the details of the reflection component. This network usually includes multiple parallel convolution paths, each path performs feature extraction at different scales to capture information of different scales, wherein the embodiment of the present invention may specifically include three scales, wherein the first scale is to directly apply a convolution operation to the input reflection component to extract local features, that is, to extract local features through a smaller convolution kernel. The second scale is to reduce the image resolution through a downsampling (such as pooling) operation, and apply a convolution operation to extract features of a larger range, and then restore it to the original resolution through an upsampling operation, that is, to process a slightly larger area through a medium-scale convolution kernel. The third scale is to extract higher-level features through a larger range of downsampling and convolution operations, and also restore the original resolution after upsampling, that is, to capture global information through a larger convolution kernel. Further, the features extracted from different scales are fused, usually using splicing or weighting or other fusion strategies to obtain a comprehensive feature map. The fused comprehensive feature map contains information of the reflection component at each scale. This step allows the network to simultaneously utilize low-scale and high-scale feature information to enhance the details and contrast of the image. Finally, the fused feature map is converted into an enhanced reflection component through a layer of convolutional generator or other generation modules. The goal of this step is to maintain the details of the reflection component and enhance its brightness and contrast.
[0085] The deep denoising autoencoder is used to remove noise from the reflection component and enhance the details of the image. Its structure includes an encoder and a decoder. The encoder part of the deep denoising autoencoder is used to map the reflection component (noisy image) to a low-dimensional latent space to extract the main features of the image and remove noise at the same time. The encoder includes a convolution layer and a pooling layer, wherein the convolution layer applies a convolution operation to the input reflection component to compress the image to a lower dimension while retaining the main features. The filter of the convolution layer can capture the features of the local area, such as edges, textures, etc. The pooling layer further reduces the spatial dimension of the image through maximum pooling or average pooling operations, extracts high-level features, thereby reducing the spatial dimension and computational complexity of the features. At this time, the input image is compressed into a low-dimensional feature representation through multiple convolution and pooling operations. This process compresses the noise into the feature space, thereby reducing the impact of the noise on the final output. Therefore, the encoder uses convolution layers and pooling layers to compress the image to a low-dimensional feature representation, while removing noise and retaining the main features. At this time, the low-dimensional feature representation (latent space) output by the encoder contains the main feature information of the enhanced reflection component and reduces the influence of noise. The noise is removed or compressed as much as possible in this process. The decoder part of the deep denoising autoencoder is used to recover a high-quality image from the latent space, that is, the denoised image. The decoder includes a deconvolution layer, an activation function and an output layer. The deconvolution layer uses an upsampling operation to gradually restore the spatial resolution of the image, and refines the details of the image through a convolution operation, so that the low-dimensional feature representation is restored to the spatial resolution of the original image, thereby reconstructing the spatial structure of the image. The activation function usually uses a nonlinear activation function (such as ReLU or Sigmoid) to ensure that the restored image has good contrast and brightness. The output layer is usually a convolution layer, and the number of channels of the image is set to the number of channels of the target image (for example, 1 for grayscale images and 3 for RGB images). At this time, the image generated by the decoder is the denoised and enhanced reflection component, with higher detail clarity and less noise. Furthermore, the loss function and optimizer are defined, where MSE (Mean Squared Error) is used as the loss function to measure the difference between the reconstructed image and the original clean image. Specifically, MSE calculates the square difference between the predicted value and the true value of each pixel in the image. At this time, the training goal is to minimize MSE so that the reconstructed image is as close to the clean image as possible. At the same time, an optimizer (such as Adam or SGD) is used to update the weights and biases of the network to minimize the loss function. At this time, the optimizer adjusts the network parameters through the gradient descent method to gradually improve the model.Furthermore, the model training of the deep denoising autoencoder is to take the noisy image as input data, pass it into the encoder part of the deep denoising autoencoder, generate the denoised image through the decoder, calculate the loss between it and the original clean image, calculate the gradient through the back propagation algorithm, and use the optimizer to update the weight of the network to reduce the value of the loss function. At this time, the trained deep denoising autoencoder can process new noisy images and generate high-quality denoised images. Its trained deep denoising autoencoder can effectively remove noise while retaining the details and structural information of the image.
[0086] Among them, since the multi-scale convolutional network can extract detailed information of the image at different scales, through feature fusion, it ensures that various details in the reflection component are preserved and enhanced. The deep denoising autoencoder effectively eliminates the noise in the reflection component through encoding-decoding, making the image cleaner and smoother while enhancing. Compared with the traditional method, the reflection enhancement method of the reflection component based on the multi-scale convolutional network and the deep denoising autoencoder can not only enhance the details in the reflection component, but also avoid over-sharpening or blurring, thereby improving the image quality during the enhancement process. At the same time, the combination of the multi-scale convolutional network and the deep denoising autoencoder can dynamically process according to the actual information in the image, making the enhancement effect more natural and adaptable to different lighting conditions. At the same time, under low-light conditions, the deep denoising autoencoder can effectively cope with various complex lighting environments, enhancing the reflection component while maintaining the naturalness and consistency of the image.
[0087] Step S40, fusing the corrected illumination component and the enhanced reflection component to obtain target image data;
[0088] In one embodiment of the present invention, the step of fusing the corrected illumination component and the enhanced reflection component to obtain the target image data includes:
[0089] Normalizing the corrected illumination component and the enhanced reflection component;
[0090] Multiply the normalized illumination component and reflection component pixel by pixel to generate target image data;
[0091] Perform contrast adjustment and color correction on the target image data.
[0092] Specifically, first, the corrected illumination component and the enhanced reflection component are normalized. The purpose of normalization is to limit the pixel values of the image to a specific range (usually 0,1) so that numerical stability can be maintained in subsequent calculations. After normalization, the illumination component and the reflection component are multiplied pixel by pixel to generate the final target image. It is also necessary to ensure that the values of the synthesized image data are within an appropriate range (such as 0 to 255) to prevent numerical overflow or under-counting due to multiplication. If necessary, apply clipping or normalization. If the original image is a color image, the color space problem needs to be handled during the synthesis process. Ensure that the reflection component and the illumination component use the same color space when synthesizing. The generated target image may have insufficient contrast, so the image needs to be contrast adjusted. Contrast enhancement can be performed by contrast stretching or using histogram equalization. The purpose of color correction is to make the synthesized image have more natural colors, and the overall color balance of the image can be adjusted by a white balance algorithm.
[0093] In summary, the image processing method for mining lamps in the above embodiments of the present invention can effectively enhance the image brightness and detail clarity in low-light environments by separately processing the illumination component and the reflection component, improve the problems of uneven illumination and insufficient brightness, and make the key details in the mine environment more clearly visible; by correcting the illumination component using a deep generative model, the illumination effect can be dynamically adjusted to reduce the uneven illumination phenomenon in the collected image caused by the complex environment of the mine; the reflection component is enhanced by a multi-scale convolutional network and a deep denoising autoencoder, and the noise and artifacts caused by interference such as light reflection and mine dust are effectively removed, so that the final generated image is cleaner, the details are richer, and the image contrast is higher; through the enhanced image data, the imaging quality of the camera in the mining lamp can be improved, so that the monitoring system can more clearly and accurately reflect the internal environment of the mine and the mining operation, effectively ensure the safety management and operation monitoring of the mine, and solve the problem of poor image quality in the existing low-light environment.
[0094] Embodiment 2
[0095] See also Figure 2 , is a schematic diagram of the structure of an image processing system for a miner's lamp provided by a second embodiment of the present invention. For the sake of convenience, only the part related to the embodiment of the present invention is shown. The image processing system for a miner's lamp includes:
[0096] The data processing module 11 is used to collect image data of the current environment and perform image decomposition on the collected image data to obtain illumination components and reflection components;
[0097] An illumination correction module 12, configured to perform illumination correction on the illumination component according to the depth generation model, and generate a corrected illumination component;
[0098] A reflection enhancement module 13, used to enhance the reflection component according to the multi-scale convolutional network and the deep denoising autoencoder to generate an enhanced reflection component;
[0099] The data fusion module 14 is used to fuse the corrected illumination component and the enhanced reflection component to obtain target image data.
[0100] Furthermore, in one embodiment of the present invention, the data processing module 11 includes:
[0101] A data conversion unit, used for converting the collected current image data into a grayscale image, and performing logarithmic transformation on the grayscale image to obtain a logarithmic domain image;
[0102] an illumination component calculation unit, configured to perform Gaussian filtering on the logarithmic domain image after logarithmic transformation to obtain a logarithmic value of the estimated illumination component, and to convert the estimated illumination component from the logarithmic domain back to the linear domain;
[0103] The reflection component calculation unit is used to calculate the reflection component according to the gray value of the current image data and the estimated illumination component.
[0104] Furthermore, in one embodiment of the present invention, the illumination correction module 12 includes:
[0105] A module building unit for building a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component;
[0106] The data definition unit is used to define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator;
[0107] The model training unit is used to train the deep generative model. In each training iteration, the generator generates a new illumination component, the discriminator evaluates the authenticity of the generated illumination component, calculates the loss of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer;
[0108] An illumination component generating unit, used for inputting the illumination component into the trained generator to generate a corrected illumination component;
[0109] The illumination component adjustment unit is used to adjust the contrast and brightness of the generated illumination components.
[0110] Further, in one embodiment of the present invention, the reflection enhancement module 13 includes:
[0111] A reflection component enhancement unit, used for extracting features of different scales in the reflection component according to a multi-scale convolutional network, fusing the features extracted from different scales to obtain a comprehensive feature map, and converting the fused comprehensive feature map into an enhanced reflection component;
[0112] The reflection component denoising unit is used to compress the reflection component enhanced by the multi-scale convolutional network into a low-dimensional feature representation and denoise it according to the encoder in the deep denoising autoencoder, and to restore the enhanced reflection component from the low-dimensional feature representation according to the decoder in the deep denoising autoencoder.
[0113] Furthermore, in one embodiment of the present invention, the data fusion module 14 includes:
[0114] A data processing unit, used for normalizing the corrected illumination component and the enhanced reflection component;
[0115] A data generating unit, used for performing pixel-by-pixel multiplication of the normalized illumination component and the reflection component to generate target image data;
[0116] The data adjustment unit is used to perform contrast adjustment and color correction on the target image data.
[0117] The image processing system for mining lamps provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.
[0118] Embodiment 3
[0119] Another aspect of the present invention is to provide a mining lamp, see Figure 3 , shown is a mining lamp in the third embodiment of the present invention, including a memory 20, a processor 10, and a program 30 stored in the memory 20 and executable on the processor. When the processor 10 executes the program 30, the image processing method for the mining lamp as in the above embodiment is implemented.
[0120] In some embodiments, the processor 10 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 20, such as executing access restriction programs.
[0121] The memory 20 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of the miner's lamp, such as the hard disk of the miner's lamp. In other embodiments, the memory 20 can also be an external storage device of the miner's lamp, such as a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. equipped on the miner's lamp. Further, the memory 20 can also include both the internal storage unit of the miner's lamp and the external storage device. The memory 20 can be used not only to store application software and various types of data installed in the miner's lamp, but also to temporarily store data that has been output or is to be output.
[0122] It should be pointed out that Figure 3 The structure shown does not constitute a limitation on the mining lamp. In other embodiments, the mining lamp may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.
[0123] In summary, the mining lamp in the above-mentioned embodiment of the present invention can effectively enhance the image brightness and detail clarity in a low-light environment by separately processing the illumination component and the reflection component, improve the problems of uneven illumination and insufficient brightness, and make the key details in the mine environment more clearly visible; by correcting the illumination component using a deep generative model, the illumination effect can be dynamically adjusted to reduce the uneven illumination phenomenon in the collected image caused by the complex environment of the mine; the reflection component is enhanced by a multi-scale convolutional network and a deep denoising autoencoder, and the noise and artifacts caused by interference such as light reflection and mine dust are effectively removed, so that the final generated image is cleaner, the details are richer, and the image contrast is higher; through the enhanced image data, the imaging quality of the camera in the mining lamp can be improved, so that the monitoring system can more clearly and accurately reflect the internal environment of the mine and the mining operation conditions, effectively ensure the safety management and operation monitoring of the mine, and solve the problem of poor image quality in the existing low-light environment.
[0124] The embodiment of the present invention further provides a storage medium on which a program is stored. When the program is executed by a processor, the image processing method for a mining lamp as described in the above embodiment is implemented.
[0125] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional units or modules as needed, that is, the internal structure of the storage device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the implementation method can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application.
[0126] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any storage medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, a "storage medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0127] More specific examples of readable storage media (a non-exhaustive list) include the following: an electrical connection with one or more wirings (electronic devices), a portable computer disk case (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the storage medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or processing in another suitable manner as necessary, and then stored in a memory.
[0128] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0129] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0130] The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.
Claims
1. An image processing method for a mining lamp, characterized in that: The method comprises: Collect image data of the current environment, and perform image decomposition on the collected image data to obtain illumination component and reflection component; Performing illumination correction on the illumination component according to the deep generative model to generate a corrected illumination component; Performing reflection enhancement on the reflection component according to a multi-scale convolutional network and a deep denoising autoencoder to generate an enhanced reflection component; The corrected illumination component and the enhanced reflection component are fused to obtain the target image data.
2. The image processing method for a mining lamp according to claim 1, characterized in that: The step of performing image decomposition on the collected image data to obtain illumination components and reflection components comprises: The collected current image data is converted into a grayscale image, and the grayscale image is logarithmically transformed to obtain a logarithmic domain image; Gaussian filtering is performed on the logarithmic domain image after logarithmic transformation to obtain the logarithmic value of the estimated illumination component, and the estimated illumination component is converted from the logarithmic domain back to the linear domain; The reflection component is calculated based on the grayscale value of the current image data and the estimated illumination component.
3. The image processing method for a mining lamp according to claim 1, characterized in that: The step of performing illumination correction on the illumination component according to the deep generative model to generate the corrected illumination component comprises: Constructing a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component; Define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator; Train a deep generative model. In each training iteration, the generator generates new illumination components, the discriminator evaluates the authenticity of the generated illumination components, calculates the losses of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer. Input the illumination component into the trained generator to generate a corrected illumination component; Applies contrast and brightness adjustments to the resulting lighting components.
4. The image processing method for a mining lamp according to claim 1, characterized in that: The step of performing reflection enhancement on the reflection component according to the multi-scale convolutional network and the deep denoising autoencoder to generate the enhanced reflection component comprises: Extract features of different scales in the reflection component according to the multi-scale convolutional network, fuse the features extracted from different scales to obtain a comprehensive feature map, and convert the fused comprehensive feature map into an enhanced reflection component; According to the encoder in the deep denoising autoencoder, the reflected component enhanced by the multi-scale convolutional network is compressed into a low-dimensional feature representation and denoised, and according to the decoder in the deep denoising autoencoder, the enhanced reflected component is restored from the low-dimensional feature representation.
5. The image processing method for a mining lamp according to claim 1, characterized in that: The step of fusing the corrected illumination component and the enhanced reflection component to obtain target image data comprises: Normalizing the corrected illumination component and the enhanced reflection component; Multiply the normalized illumination component and reflection component pixel by pixel to generate target image data; Perform contrast adjustment and color correction on the target image data.
6. An image processing system for a mining lamp, characterized in that: The system comprises: A data processing module is used to collect image data of the current environment and perform image decomposition on the collected image data to obtain illumination components and reflection components; An illumination correction module, used for performing illumination correction on the illumination component according to the deep generation model to generate a corrected illumination component; A reflection enhancement module, used to enhance the reflection component according to a multi-scale convolutional network and a deep denoising autoencoder to generate an enhanced reflection component; The data fusion module is used to fuse the corrected illumination component and the enhanced reflection component to obtain the target image data.
7. The image processing system for a mining lamp according to claim 6, characterized in that: The data processing module comprises: A data conversion unit, used for converting the collected current image data into a grayscale image, and performing logarithmic transformation on the grayscale image to obtain a logarithmic domain image; an illumination component calculation unit, configured to perform Gaussian filtering on the logarithmic domain image after logarithmic transformation to obtain a logarithmic value of the estimated illumination component, and to convert the estimated illumination component from the logarithmic domain back to the linear domain; The reflection component calculation unit is used to calculate the reflection component according to the gray value of the current image data and the estimated illumination component.
8. The image processing system for a mining lamp according to claim 6, characterized in that: The illumination correction module comprises: A module building unit for building a deep generative model including a generator and a discriminator network, wherein the generator is used to generate a corrected illumination component and the discriminator is used to evaluate the generated illumination component; The data definition unit is used to define the loss function and optimizer in the deep generative model, where the loss function is used to guide the training process and the optimizer is used to update the weights of the generator and discriminator; The model training unit is used to train the deep generative model. In each training iteration, the generator generates a new illumination component, the discriminator evaluates the authenticity of the generated illumination component, calculates the loss of the generator and the discriminator, and updates the weights of the generator and the discriminator through the optimizer; An illumination component generating unit, used for inputting the illumination component into the trained generator to generate a corrected illumination component; The illumination component adjustment unit is used to adjust the contrast and brightness of the generated illumination components.
9. A storage medium storing a program, characterized in that: When the program is executed by a processor, the image processing method for a mining lamp according to any one of claims 1 to 5 is implemented.
10. A mining lamp, characterized in that: The invention comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the image processing method for a mining lamp according to any one of claims 1 to 5 is implemented.
Citation Information
Cited By
Visual identification system for intelligent electric appliance
CN119942083A
Drill rod counting method and device applied to underground coal mine low illumination
CN120877210A
