Intelligent storage shelf detection method and system based on artificial intelligence

Abstract

The invention provides an intelligent storage shelf detection method and system based on artificial intelligence, and relates to the field of target detection, and the method comprises the steps: obtaining a sample storage shelf image and a to-be-detected storage shelf image; the sample storage shelf images comprise target goods with different illumination, different placement modes and different shielding areas; performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; performing chromatic aberration balance on the first pre-processed image to obtain a second pre-processed image; inputting the second pre-processed image into the initial detection model to obtain an intelligent storage shelf detection model; and inputting the to-be-detected storage shelf image into the intelligent storage shelf detection model for detection to obtain a detection result. According to the method, the environmental change sensitivity is effectively reduced, and the recognition precision and reliability in practical application are improved.

Description

Technical Field

[0001] The present invention relates to the field of target detection, and in particular to an intelligent storage shelf detection method and system based on artificial intelligence. Background Art

[0002] With the rapid development of e-commerce and logistics industries, warehouse management faces multiple challenges such as inventory management, cargo tracking, space utilization, etc. Traditional manual management methods are inefficient, prone to errors, and cannot meet the needs of modern warehousing.

[0003] Advances in artificial intelligence (AI) technology have provided new solutions for warehouse management. Through machine learning, computer vision and other technologies, automatic detection and management of shelves can be achieved, improving warehouse efficiency and accuracy. Computer vision technology can monitor shelf status in real time through cameras and identify the type, quantity and location of goods. This technology can quickly process large amounts of image data and promptly detect problems such as out-of-stock and misplaced items on shelves.

[0004] The invention patent with application number 202111389358.4 discloses an intelligent cargo supervision method based on a twin neural network, including: obtaining a shelf image to be detected; the shelf image to be detected is based on an image of each shelf in the warehouse taken by an industrial camera; searching in the standard shelf image library to obtain a standard shelf image corresponding to the shelf image to be detected; the standard shelf image is a full warehouse state, and the display of goods meets the set requirements; according to the shelf image to be detected and the corresponding standard shelf image, a trained cargo recognition network model is used to determine the detection result of the shelf image to be detected; the detection results include: the inventory and location of each type of goods, the number of free positions on the shelf, and whether the goods meet the storage placement requirements; according to the shelf image to be detected and the corresponding standard shelf image, the process of using the trained cargo recognition network model to determine the detection result of the shelf image to be detected is: using the ResNet50 network in the trained cargo recognition network model to extract the cargo features of the shelf image to be detected and the corresponding standard shelf image; using the feature pyramid network FPN in the trained cargo recognition network model to perform cargo feature extraction The feature fusion and hierarchical detection operations of the features are performed; the region proposal network and ROIPooling layer in the trained goods recognition network model are used in turn to output a series of candidate region feature maps of the same size; the candidate region feature maps are respectively input into two fully connected layers for feature extraction and classification, and then respectively input into the regression branch and the classification branch to determine the category of goods to which the candidate region feature maps belong and the location information on the shelf; according to the candidate region feature maps and the corresponding location information, the CNet network is used to determine the storage location code of each commodity on the shelf in the shelf image to be detected; the storage location code is used to determine whether the goods meet the storage placement requirements; the CNet network includes: two convolution layers with a convolution kernel of 1*1 connected in sequence, a fully connected layer and a loss layer; according to the category of goods to which the candidate region feature maps belong and the location information on the shelf, the inventory of each type of goods and the number of free storage locations on the shelf are determined; according to the storage location code of each commodity on the shelf in the shelf image to be detected, whether the goods meet the storage placement requirements; the loss function value of the trained goods recognition network model is determined according to the classification loss function, the regression loss function and the joint contrast loss function.

[0005] Although the above-mentioned intelligent cargo supervision method based on twin neural networks is advanced in cargo detection and supervision, it has the defect of sensitivity to environmental changes. That is, this method is more sensitive to environmental changes (such as lighting, occlusion, changes in cargo placement, etc.), resulting in a decrease in recognition accuracy. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide an intelligent warehouse shelf detection method and system based on artificial intelligence, which effectively reduces the problem of sensitivity to environmental changes and improves its recognition accuracy and reliability in practical applications.

[0007] To achieve the above object, the present invention provides the following solutions: An intelligent storage shelf detection method based on artificial intelligence, comprising: Acquire sample storage shelf images and test storage shelf images; the sample storage shelf images contain target goods with different lighting, different placement methods, and different occlusion areas; Performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; Performing color difference balance on the first preprocessed image to obtain a second preprocessed image; Inputting the second preprocessed image into the initial detection model to obtain an intelligent storage shelf detection model; The image of the storage shelf to be tested is input into the intelligent storage shelf detection model for detection to obtain a detection result.

[0008] Preferably, performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image includes: Performing filtering processing on each of the sample storage shelf images to obtain a filtered image; Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image; Performing histogram equalization processing on the denoised image to obtain the first preprocessed image.

[0009] Preferably, filtering is performed on each of the sample storage shelf images to obtain a filtered image, including: Performing wavelet decomposition on the sample storage shelf image at multiple scales to obtain corresponding wavelet coefficients; A filtering threshold is constructed according to the size and decomposition scale of the sample storage shelf image; the formula of the filtering threshold is: ;in, Indicated in j The filtering threshold at the decomposition scale is Indicated in j The standard deviation of noise at each decomposition scale is Indicated in j The signal length at each decomposition scale is j Indicates the decomposition scale; The image filtering model is constructed using the filtering threshold; the formula of the image filtering model is: ;in,sign is the symbolic function, a is the first preset coefficient, b is the second preset coefficient, Indicated in j The decomposition scale k wavelet coefficients, represents the wavelet coefficients after filtering; The sample storage shelf image is filtered using the image filtering model to obtain the filtered image.

[0010] Preferably, constructing an image enhancement window and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image includes: Detecting the filtered image using an image recognition model to obtain a proportion value under each image enhancement window; When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising; The image enhancement window is slid until the entire filtered image is traversed to obtain the denoised image.

[0011] Preferably, the filtered image is detected using an image recognition model to obtain a proportion value under each image enhancement window, including: Using formula Get the pixel difference of each pixel under each image enhancement window; where, Represents the pixel difference of the image at position (i, j), Represents the pixel value of the image at position (i, j), Represents the pixel value of the filtered image at position (i, j); Pixels whose pixel difference is greater than the feature threshold are taken as feature points and the number is recorded; The ratio of the feature points to all the pixels in the image enhancement window is taken as the ratio value.

[0012] Preferably, performing color difference balancing on the first preprocessed image to obtain a second preprocessed image includes: Calculate the pixel mean of the first preprocessed image in the three RGB channels; wherein the pixel mean calculation formula is: ;in, represents the pixel mean, represents the pixel mean value of the first preprocessed image in the red channel, represents the pixel mean value of the first preprocessed image in the green channel, represents the pixel mean value of the first preprocessed image in the blue channel; Calculating a balance coefficient for each color channel according to the pixel mean; The first preprocessed image is subjected to color difference balance using the balance coefficient to obtain the second preprocessed image.

[0013] Preferably, the calculation formula of the balance coefficient is: in, represents the balance coefficient of the red channel, represents the balance coefficient of the green channel, represents the balance coefficient of the blue channel, represents the pixel value of the red channel after balance, Represents the pixel value of the green channel after balance, represents the pixel value of the blue channel after balance, represents the pixel value of the first preprocessed image in the red channel, represents the pixel value of the first preprocessed image in the green channel, Represents the pixel value of the first preprocessed image in the blue channel.

[0014] Preferably, the initial detection model includes an input network, a backbone network, a neck network and a detection head connected in sequence; the input network is used to perform data enhancement and image preprocessing operations on the input image to obtain a standard image; the backbone network is used to extract image features of the standard image based on the ELAN model and the MP model to obtain a feature map; the neck network is used to perform feature fusion and enhancement on the feature map to obtain a fusion map; the detection head is used to perform feature detection based on the fusion map.

[0015] Preferably, the backbone network of the backbone network is a MobileNetV3 network.

[0016] An intelligent storage shelf detection system based on artificial intelligence, comprising: An image acquisition module is used to acquire sample storage shelf images and test storage shelf images; the sample storage shelf images contain target goods with different lighting, different placement methods and different occlusion areas; A first preprocessing module, used for performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; A second preprocessing module, configured to perform color difference balance on the first preprocessed image to obtain a second preprocessed image; A model training module, used for inputting the second preprocessed image into an initial detection model for training to obtain an intelligent storage shelf detection model; The detection module is used to input the image of the storage shelf to be tested into the intelligent storage shelf detection model for detection to obtain a detection result.

[0017] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: The present invention provides an intelligent warehouse shelf detection method and system based on artificial intelligence, the method comprising: obtaining a sample warehouse shelf image and a warehouse shelf image to be tested; the sample warehouse shelf image contains target goods with different lighting, different placement methods and different occlusion areas; performing image filtering and image enhancement on the sample warehouse shelf image to obtain a first preprocessed image; performing color difference balance on the first preprocessed image to obtain a second preprocessed image; inputting the second preprocessed image into an initial detection model to obtain an intelligent warehouse shelf detection model; inputting the warehouse shelf image to be tested into the intelligent warehouse shelf detection model for detection to obtain a detection result. The present invention effectively reduces the problem of sensitivity to environmental changes. Relying on high-quality preprocessed data and diversified training samples, the model can more robustly cope with different lighting, occlusion and placement methods, thereby improving its recognition accuracy and reliability in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0019] Figure 1 A flow chart of a method provided by an embodiment of the present invention; Figure 2 A schematic diagram of a technical route provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The purpose of the present invention is to provide an intelligent storage shelf detection method and system based on artificial intelligence, which can lay a solid foundation for object detection in a complex and dynamic environment.

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides an intelligent storage shelf detection method based on artificial intelligence, comprising: Step 100: Acquire a sample storage shelf image and a storage shelf image to be tested; the sample storage shelf image contains target goods with different lighting, different placement methods, and different occlusion areas; Step 200: performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; Step 300: performing color difference balance on the first preprocessed image to obtain a second preprocessed image; Step 400: input the second preprocessed image into the initial detection model to obtain an intelligent storage shelf detection model; Step 500: Input the image of the storage shelf to be tested into the intelligent storage shelf detection model for detection to obtain the detection result.

[0024] Specifically, the steps of this embodiment include: 1. Acquisition of sample storage shelf images 1.1. Data acquisition equipment uses high-resolution industrial cameras or professional photography equipment to ensure that the image quality is high enough for subsequent processing and analysis. The camera should have a good dynamic range to meet the shooting needs under different lighting conditions.

[0025] 1.2. Scene setting: In an actual warehouse environment, select multiple different shelves for shooting. Each shelf should contain multiple types of goods and be arranged in different ways. For example, the goods can be placed randomly, arranged neatly, or partially blocked to simulate a real warehouse scene.

[0026] 1.3 Lighting conditions During the shooting process, adjust the lighting conditions to obtain diverse sample images. Shoot at different time periods (such as morning, noon, and evening), or use adjustable light sources (such as LED lights) to simulate different light intensities and directions. Ensure that each sample image reflects different lighting effects.

[0027] 1.4. Occlusion When shooting, some obstructions (such as other goods, packaging materials, etc.) are deliberately set up to simulate the possible obstructions in actual warehousing. The images of each obstruction are recorded for subsequent analysis and model training.

[0028] 2. Acquisition of images of storage shelves to be tested 2.1 Real-time monitoring system A real-time monitoring system is deployed in the warehouse environment, and the same type of industrial cameras are used to regularly capture images of the warehouse shelves to be tested. These images are acquired at different time periods and under different operating conditions to ensure the diversity and representativeness of the data.

[0029] 2.2. Automated acquisition The automated acquisition of the images to be tested can be achieved by setting a timed shooting or trigger mechanism (such as automatic shooting when goods enter or leave). This method can ensure that the latest shelf status images are obtained in a timely manner in actual operation.

[0030] 2.3. Data Annotation: Annotate the acquired images of the storage shelves to be tested, and record the type, location and status of the goods in each image. These annotation information will be used for subsequent model training and evaluation to ensure that the detection model can accurately identify and locate the goods.

[0031] 3. Data storage and management 3.1. Database Construction The acquired sample storage shelf images and the storage shelf images to be tested are stored in the database and managed in a structured manner. Each image is accompanied by relevant metadata (such as shooting time, lighting conditions, goods placement, etc.) for subsequent retrieval and analysis.

[0032] 3.2. Data backup and maintenance Back up the database regularly to ensure data security and integrity. At the same time, update the sample image library regularly to reflect changes in the storage environment and the introduction of new goods.

[0033] Through the above steps, this embodiment can effectively obtain a variety of sample storage shelf images and storage shelf images to be tested, providing a solid data foundation for subsequent intelligent detection model training. This embodiment not only improves the representativeness of the data, but also enhances the adaptability and robustness of the model in practical applications.

[0034] Preferably, performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image includes: Performing filtering processing on each of the sample storage shelf images to obtain a filtered image; Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image; Performing histogram equalization processing on the denoised image to obtain the first preprocessed image.

[0035] Preferably, filtering is performed on each of the sample storage shelf images to obtain a filtered image, including: Performing wavelet decomposition on the sample storage shelf image at multiple scales to obtain corresponding wavelet coefficients; A filtering threshold is constructed according to the size and decomposition scale of the sample storage shelf image; the formula of the filtering threshold is: ;in, Indicated in j The filtering threshold at the decomposition scale is Indicated in j The standard deviation of noise at each decomposition scale is Indicated in j The signal length at each decomposition scale is j Indicates the decomposition scale; The image filtering model is constructed using the filtering threshold; the formula of the image filtering model is: ;in, sign is the symbolic function, a is the first preset coefficient, b is the second preset coefficient, Indicated in j The decomposition scale k wavelet coefficients, represents the wavelet coefficients after filtering; The sample storage shelf image is filtered using the image filtering model to obtain the filtered image.

[0036] Specifically, in this embodiment, first, wavelet decomposition is performed on the sample storage shelf image. Wavelet transform can capture the local features of the image at different scales and effectively distinguish the components of different frequencies. This step of the embodiment can obtain high-frequency and low-frequency information, which are used for denoising and retaining image details respectively. According to the size and decomposition scale of the sample image, a corresponding filtering threshold is constructed. The threshold is based on the formula of the noise standard deviation and the signal length, and the filtering effect can be dynamically adjusted so that the image processing under different conditions can achieve the best denoising effect.

[0037] Furthermore, this embodiment uses the constructed filtering threshold to design an image filtering model to process the wavelet coefficients. Specifically, a sign function is used, which is combined with two preset coefficients. Through simple numerical operations, the influence of noise can be effectively reduced while retaining important image information. After the processing is completed, the obtained filtered image is denoised to remove possible noise interference. Then, histogram equalization processing is implemented to improve the contrast of the image, enhance the details and visibility of the image, and thus obtain a first preprocessed image.

[0038] Specifically, this embodiment uses wavelet decomposition and its adaptive filtering threshold to accurately distinguish between signals and noise, and has a better denoising effect than traditional filtering technology. This is particularly important for processing complex storage environment images, because these images are often affected by environmental noise. Wavelet transform can capture image details at different scales with significant effects. Compared with simple smoothing, this embodiment can better retain edge and texture information, ensure that important features are not blurred, and thus improve the recognition rate of subsequent detection models. By dynamically adjusting the filtering threshold according to the characteristics of the image, the processing process has adaptive capabilities, can effectively cope with various lighting and occlusion conditions, and enhances the adaptability of the model in complex environments. Histogram equalization processing further improves the visibility of the image, making the identification of target goods clearer, and ensuring that subsequent detection algorithms can obtain more accurate input data during processing.

[0039] In summary, through the combination of wavelet decomposition, dynamic filtering threshold construction and histogram equalization, this technology can not only effectively denoise and enhance images, but also improve the overall detection accuracy, providing a strong guarantee for intelligent warehouse shelf detection.

[0040] Preferably, constructing an image enhancement window and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image includes: Detecting the filtered image using an image recognition model to obtain a proportion value under each image enhancement window; When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising; The image enhancement window is slid until the entire filtered image is traversed to obtain the denoised image.

[0041] Preferably, the filtered image is detected using an image recognition model to obtain a proportion value under each image enhancement window, including: Using formula Get the pixel difference of each pixel under each image enhancement window; where, Represents the pixel difference of the image at position (i, j), Represents the pixel value of the image at position (i, j), Represents the pixel value of the filtered image at position (i, j); Pixels whose pixel difference is greater than the feature threshold are taken as feature points and the number is recorded; The ratio of the feature points to all the pixels in the image enhancement window is taken as the ratio value.

[0042] Specifically, the present embodiment first sets an image enhancement window of appropriate size, which is slid over the filtered image. The size of the window is selected according to the features of the specific image and the size of the target object to ensure that sufficient detail information can be captured. At each window position, the pixel difference of each pixel in the image is calculated using a formula. By comparing the pixel value of the current window with the corresponding pixel value of the filtered image, possible feature changes and texture information in the image are detected. Subsequently, the present embodiment compares the calculated pixel difference with the set feature threshold, identifies and records the pixels in the window whose pixel difference is greater than the feature threshold, and counts the number of these feature points.

[0043] Furthermore, this embodiment compares the number of feature points in the current window with the total number of pixels in the window, and calculates the ratio of the feature points to reflect the density of the associated features in the window area. When the ratio of a window exceeds a preset threshold, it indicates that there may be significant features in the area. Median denoising is performed on this area, and median filtering can effectively remove noise while retaining edges and important details, thereby obtaining an image after median denoising. The sliding image enhancement window continues to traverse the entire filtered image until a complete denoised image is obtained. This process can achieve denoising in a larger area while maintaining the overall structure and features of the image.

[0044] Furthermore, this embodiment, through pixel difference calculation and feature threshold setting, can effectively identify important feature points on the image and ensure that the features are retained during the denoising process. This is particularly important for storage shelf detection, and the retention of detailed features has a great impact on subsequent recognition work. The necessity of denoising is judged based on the proportion of feature points, so that this method has adaptive characteristics. Median denoising in areas with dense feature points can effectively remove noise and ensure the retention of important features, thereby enhancing the robustness of the model in complex and changing environments. The application of median filtering can effectively suppress high-frequency noise, while better protecting edge details, avoiding image distortion and artifact problems caused by traditional mean filtering, and improving image quality. This embodiment can greatly improve the accuracy and effectiveness of subsequent intelligent detection models by further analyzing and processing the denoised image, ensuring that higher recognition accuracy is achieved in practical applications.

[0045] In summary, this embodiment implements an effective denoising scheme by constructing an image enhancement window and performing feature point analysis, which not only improves the image quality, but also enhances the adaptability of subsequent detection models in complex environments, and has significant application value.

[0046] Preferably, performing color difference balancing on the first preprocessed image to obtain a second preprocessed image includes: Calculate the pixel mean of the first preprocessed image in the three RGB channels; wherein the pixel mean calculation formula is: ;in, represents the pixel mean, represents the pixel mean value of the first preprocessed image in the red channel, represents the pixel mean value of the first preprocessed image in the green channel, represents the pixel mean value of the first preprocessed image in the blue channel; Calculating a balance coefficient for each color channel according to the pixel mean; The first preprocessed image is subjected to color difference balance using the balance coefficient to obtain the second preprocessed image.

[0047] Preferably, the calculation formula of the balance coefficient is: in, represents the balance coefficient of the red channel, represents the balance coefficient of the green channel, represents the balance coefficient of the blue channel, represents the pixel value of the red channel after balance, Represents the pixel value of the green channel after balance, represents the pixel value of the blue channel after balance, represents the pixel value of the first preprocessed image in the red channel, represents the pixel value of the first preprocessed image in the green channel, Represents the pixel value of the first preprocessed image in the blue channel.

[0048] Specifically, this embodiment performs equalization processing on each color channel. This technology can effectively eliminate color deviations caused by changes in lighting or different shooting conditions, and ensure the color consistency of images in different environments. This is crucial for subsequent image analysis and object detection. Color balance processing can enhance the overall visual effect of the image, making the target object in the image more vivid and prominent. This enhancement effect helps to improve the recognition ability of the detection model in complex environments, especially in scenes with uneven lighting or rich colors. By eliminating color differences, this embodiment can maintain a high recognition accuracy when processing images under different conditions. This robustness enables the intelligent storage shelf detection system to better adapt to various environmental changes in practical applications, reducing false detections and missed detections. And the image after color balance provides a more stable foundation for subsequent feature extraction and analysis, simplifying the complexity of subsequent processing. The model can focus more on identifying and locating targets without worrying about interference caused by color differences.

[0049] In summary, this embodiment performs color balance processing on the first preprocessed image. This technology not only improves the color consistency and visual effect of the image, but also enhances the robustness of the intelligent detection model, providing more reliable input data for subsequent object detection.

[0050] Preferably, the initial detection model includes an input network, a backbone network, a neck network and a detection head connected in sequence; the input network is used to perform data enhancement and image preprocessing operations on the input image to obtain a standard image; the backbone network is used to extract image features of the standard image based on the ELAN model and the MP model to obtain a feature map; the neck network is used to perform feature fusion and enhancement on the feature map to obtain a fusion map; the detection head is used to perform feature detection based on the fusion map.

[0051] Preferably, the backbone network of the backbone network is a MobileNetV3 network.

[0052] Specifically, this embodiment first performs data enhancement and preprocessing operations on the input image through the input network. The purpose of this stage is to standardize the original image to improve the robustness and adaptability of the model, including operations such as cropping, rotation, scaling and color transformation of the image. After processing, a standard image is obtained, and these standardized data are more conducive to subsequent feature extraction. Next, MobileNetV3 is used as the backbone network to extract features from the standard image. MobileNetV3 is a lightweight convolutional neural network optimized for mobile devices and edge computing. It has a small model size and low computational complexity, and can effectively reduce the inference time. The network combines the ELAN model and the MP model to extract rich image features and form a feature map. Then, the neck network performs feature fusion and enhancement on the extracted feature map. In this step, multiple feature maps are fused by upsampling or concatenation to capture multi-scale information and contextual information. This fusion helps to improve the accuracy of object detection because it combines features from different levels and enhances the model's ability to detect small targets. Finally, the detection head performs feature detection based on the fusion map, and finally outputs the detection results, including the category and position of the target, through classification and regression operations. This process ensures that the model can not only recognize the existence of an object, but also accurately locate it.

[0053] Furthermore, the use of the MobileNetV3 backbone network used in this embodiment enables the model to achieve a good balance between performance and computational efficiency. Its lightweight design enables the model to maintain efficient operation in resource-constrained environments (such as mobile devices or embedded systems), which is suitable for practical applications. By introducing the ELAN model and the MP model, the backbone network can extract more rich features, especially in complex backgrounds and diverse target scenes, significantly improving the accuracy of target detection. This feature extraction capability provides a reliable foundation for subsequent fusion and detection. The data enhancement and preprocessing process of the input network improves the adaptability of the model to various input images. By dynamically adjusting the data, the algorithm can still show good detection capabilities when facing different lighting, placement and occlusion conditions. The feature fusion design of the neck network ensures that the model can make full use of features at different levels and scales, enhance the detection capability of small objects and partially obscured objects, thereby improving the overall detection effect.

[0054] In summary, the detection model of this embodiment ensures that the intelligent target detection system can complete tasks efficiently and accurately in complex warehousing environments through reasonable network architecture design and efficient feature extraction process, providing strong technical support for further applications.

[0055] Corresponding to the above method, such as Figure 2 As shown, this embodiment also provides an intelligent storage shelf detection system based on artificial intelligence, including: An image acquisition module is used to acquire sample storage shelf images and test storage shelf images; the sample storage shelf images contain target goods with different lighting, different placement methods and different occlusion areas; A first preprocessing module, used for performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; A second preprocessing module, configured to perform color difference balance on the first preprocessed image to obtain a second preprocessed image; A model training module, used for inputting the second preprocessed image into an initial detection model for training to obtain an intelligent storage shelf detection model; The detection module is used to input the image of the storage shelf to be tested into the intelligent storage shelf detection model for detection to obtain a detection result.

[0056] The beneficial effects of the present invention are as follows: The intelligent storage shelf detection method and system of the present invention effectively reduces the problem of sensitivity to environmental changes. Relying on high-quality preprocessed data and diversified training samples, the model can more robustly cope with different lighting, occlusion and placement methods, improving its recognition accuracy and reliability in practical applications. This method lays a solid foundation for object detection in complex and dynamic environments.

[0057] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0058] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An intelligent storage shelf detection method based on artificial intelligence, characterized in that: include: Acquire sample storage shelf images and test storage shelf images; the sample storage shelf images contain target goods with different lighting, different placement methods, and different occlusion areas; Performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; Performing color difference balance on the first preprocessed image to obtain a second preprocessed image; Inputting the second preprocessed image into the initial detection model to obtain an intelligent storage shelf detection model; The image of the storage shelf to be tested is input into the intelligent storage shelf detection model for detection to obtain a detection result.

2. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 1 is characterized in that: Performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image includes: Performing filtering processing on each of the sample storage shelf images to obtain a filtered image; Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image; Performing histogram equalization processing on the denoised image to obtain the first preprocessed image.

3. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 2 is characterized in that: Performing filtering processing on each of the sample storage shelf images to obtain a filtered image includes: Performing wavelet decomposition on the sample storage shelf image at multiple scales to obtain corresponding wavelet coefficients; A filtering threshold is constructed according to the size and decomposition scale of the sample storage shelf image; the formula of the filtering threshold is: ,in, Indicated in j The filtering threshold at the decomposition scale is Indicated in j The standard deviation of noise at each decomposition scale is Indicated in j The signal length at each decomposition scale is j Indicates the decomposition scale; The image filtering model is constructed using the filtering threshold; the formula of the image filtering model is: ;in, sign is the symbolic function, a is the first preset coefficient, b is the second preset coefficient, Indicated in j The decomposition scale k wavelet coefficients, represents the wavelet coefficients after filtering; The sample storage shelf image is filtered using the image filtering model to obtain the filtered image.

4. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 3 is characterized in that: Constructing an image enhancement window, and using the image enhancement window to traverse the filtered image to perform denoising to obtain a denoised image, including: Detecting the filtered image using an image recognition model to obtain a proportion value under each image enhancement window; When the proportion value in the image enhancement window is greater than a preset threshold, median denoising is performed on the image in the corresponding image enhancement window to obtain an image after median denoising; The image enhancement window is slid until the entire filtered image is traversed to obtain the denoised image.

5. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 4 is characterized in that: The filtered image is detected using an image recognition model to obtain a proportion value under each image enhancement window, including: Using formula Get the pixel difference of each pixel under each image enhancement window; where, Represents the pixel difference of the image at position (i, j), Represents the pixel value of the image at position (i, j), Represents the pixel value of the filtered image at position (i, j); Pixels whose pixel difference is greater than the feature threshold are taken as feature points and the number is recorded; The ratio of the feature points to all the pixels in the image enhancement window is taken as the ratio value.

6. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 1 is characterized in that: Performing color difference balancing on the first preprocessed image to obtain a second preprocessed image includes: Calculate the pixel mean of the first preprocessed image in the three RGB channels; wherein the pixel mean calculation formula is: ;in, represents the pixel mean, represents the pixel mean value of the first preprocessed image in the red channel, represents the pixel mean value of the first preprocessed image in the green channel, represents the pixel mean value of the first preprocessed image in the blue channel; Calculating a balance coefficient for each color channel according to the pixel mean; The first preprocessed image is subjected to color difference balance using the balance coefficient to obtain the second preprocessed image.

7. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 6 is characterized in that: The calculation formula of the balance coefficient is: in, represents the balance coefficient of the red channel, represents the balance coefficient of the red channel, represents the balance coefficient of the blue channel, represents the pixel value of the red channel after balance, Represents the pixel value of the green channel after balancing, represents the pixel value of the blue channel after balance, represents the pixel value of the first preprocessed image in the red channel, represents the pixel value of the first preprocessed image in the green channel, Represents the pixel value of the first preprocessed image in the blue channel.

8. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 1 is characterized in that: The initial detection model includes an input network, a backbone network, a neck network and a detection head connected in sequence; the input network is used to perform data enhancement and image preprocessing operations on the input image to obtain a standard image; the backbone network is used to extract image features of the standard image based on the ELAN model and the MP model to obtain a feature map; the neck network is used to perform feature fusion and enhancement on the feature map to obtain a fusion map; the detection head is used to perform feature detection based on the fusion map.

9. The method for detecting intelligent storage shelves based on artificial intelligence according to claim 8 is characterized in that: The backbone network of the backbone network is the MobileNetV3 network.

10. An intelligent storage shelf detection system based on artificial intelligence, characterized in that: include: An image acquisition module is used to acquire sample storage shelf images and images of storage shelves to be tested; the sample storage shelf images contain target goods with different lighting, different placement methods and different occlusion areas; A first preprocessing module, used for performing image filtering and image enhancement on the sample storage shelf image to obtain a first preprocessed image; A second preprocessing module, configured to perform color difference balance on the first preprocessed image to obtain a second preprocessed image; A model training module, used for inputting the second preprocessed image into an initial detection model for training to obtain an intelligent storage shelf detection model; The detection module is used to input the image of the storage shelf to be tested into the intelligent storage shelf detection model for detection to obtain a detection result.

Images