Shared device scheduling method, device, terminal and storage medium

Abstract

The present invention provides a method, apparatus, terminal, and storage medium for scheduling shared equipment, including: acquiring and classifying real-time images of a target area; and, for target area images classified as point-stacked images, counting target equipment and determining whether the target area requires scheduling based on the count results. The present invention enables automatic identification of shared equipment scheduling. By classifying real-time images and counting specific equipment within the images, it accurately, effectively, and efficiently analyzes the images in real time, compensating for the inadequacy of offline staff in creating their own scheduling tasks, saving the labor costs of manual approvals, and avoiding the cost losses to the company caused by the complexity of offline personnel. This is an important technical solution for risk control in the shared industry.

Description

Technical Field

[0001] The present invention relates to the technical field of device scheduling, and in particular to a method, device, terminal and storage medium for scheduling shared devices. Background Art

[0002] Sharing equipment is an extremely asset-heavy business sector, with millions of shared vehicles deployed on roads. Refined, intelligent, and efficient operations are key competitive advantages in this sector. Currently, it's widely accepted that the future of the sharing industry lies in combining intelligent online algorithmic decision-making with efficient offline operations and maintenance.

[0003] Due to their operational characteristics, shared equipment is prone to bulk accumulation at certain times and locations, such as at subway station entrances during rush hour. This not only affects the vehicles' secondary production orders, but also causes congestion and affects the cityscape. Offline operations and maintenance must promptly dispatch such accumulations. However, due to the complexity of offline operations personnel, overcorrection is inevitable, with operations and maintenance dispatching vehicles that do not need to be dispatched, resulting in a large amount of ineffective costs. In response to this situation, how to accurately and efficiently screen the large number of work orders generated daily, filter out scenarios that do not require dispatch, save costs, and allocate transportation capacity to more important tasks is a key part of achieving refined operations for shared equipment. Summary of the Invention

[0004] In view of the above-mentioned defects of the prior art, the present invention provides a shared equipment scheduling method, device, terminal and storage medium to solve technical problems such as how to accurately and efficiently screen the large number of work orders generated daily and filter out scenarios that do not require scheduling.

[0005] To achieve the above-mentioned objectives, the present invention provides a shared device scheduling method, comprising: obtaining real-time collected target area images and classifying them; for target area images classified as point accumulation images, counting target devices and determining whether the target area needs to be scheduled based on the counting results.

[0006] In a preferred embodiment of the present invention, the target area image is further classified using a deep learning neural network model.

[0007] In another preferred embodiment of the present invention, the process of classifying the target area image by the deep learning neural network model includes: acquiring multiple historically collected images and performing preliminary scene classification based on image granularity features; annotating the historically collected images according to the preliminary scene classification results to form a training set, which is input into the deep learning neural network model and then trained to obtain a scene classification model; inputting the target area image collected in real time into the scene classification model for prediction to obtain the scene classification result output by the scene classification model.

[0008] In another preferred embodiment of the present invention, the deep learning neural network model includes an input layer, a convolution layer, a pooling layer, a batch normalization layer and an output layer; the training process of the deep learning neural network model includes: inputting the point map of the historical acquisition image into the input layer; using the convolution layer to perform convolution calculation on the point map, using the pooling layer to compress the convolution calculation result, and then using the batch normalization layer to perform batch normalization processing on the data compression result to obtain an image feature map; in the output layer, using the same number of output neurons as the number of scene categories to encode and extract the image feature map to obtain a logits vector, and after inputting the logits vector into the softmax function, the probability that the point map of the historical acquisition image belongs to each scene category is obtained.

[0009] In another preferred embodiment of the present invention, the deep learning neural network model uses a sample-biased loss function during training to compensate for the imbalance of sample categories by setting corresponding weights.

[0010] In another preferred embodiment of the present invention, the process of counting target devices for the target area image of the classified point accumulation image includes: performing target detection on the target area image using a target detection model to obtain prediction boxes of all target positions; converting the target area image into an HSI image, and selecting a color range to characterize the target device based on the HSI image; judging whether there is a false detection based on whether there is an intersection area between the prediction box of the target position and the image mask of the color range, and counting after eliminating the target devices judged to be falsely detected.

[0011] In another preferred embodiment of the present invention, a target detection model is used to perform target detection on the target area image to obtain a prediction frame of the target position, including: selecting the optimal prediction frame from multiple prediction frames corresponding to a target position and removing redundant prediction frames to obtain a prediction frame uniquely corresponding to the target position.

[0012] To achieve the above-mentioned objectives, the present invention provides a shared device scheduling device, including: a scene classification module, used to obtain and classify target area images collected in real time; a scheduling allocation module, used to count target devices for target area images classified as point accumulation images and determine whether the target area needs to be scheduled based on the counting results.

[0013] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the shared device scheduling method when executed by a processor.

[0014] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides an electronic terminal, including: a processor and a memory; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory so that the terminal executes the shared device scheduling method.

[0015] The shared equipment scheduling method, device, terminal and storage medium provided by the present invention have the following technical effects: the present invention realizes the automatic identification of shared equipment scheduling, and accurately, in real time and efficiently completes the real-time analysis of the image by classifying the real-time image and counting the specific equipment in the image, which makes up for the lack of ability of offline staff in self-building scheduling tasks, saves the manpower cost of manual approval, and avoids the cost loss to the company caused by the complexity of the offline personnel composition. It is an important technical solution in the field of risk control in the sharing industry.

[0016] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a shared device scheduling method in an embodiment of the present invention.

[0018] Figure 2 It is a schematic diagram of the process of classifying target area images using a deep learning neural network model in an embodiment of the present invention.

[0019] Figure 3 It is a structural diagram of an electronic terminal in an embodiment of the present invention.

[0020] Figure 4 It is a structural diagram of a shared device scheduling apparatus in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0022] While some exemplary embodiments of the present invention have been described for purposes of illustration, it should be understood that the present invention may be implemented in other ways not specifically shown in the drawings.

[0023] <1> Deep learning: A key research area within machine learning, this approach focuses on studying the inherent patterns and representational hierarchies of sample data. The information gained from this learning process significantly aids in interpreting data such as text, images, and sound. The ultimate goal of machine learning is to enable machines to develop human-like analytical learning capabilities, enabling them to recognize text, images, and sound.

[0024] <2> Neural network: It is a complex network system formed by a large number of simple processing units (neurons) that are widely interconnected. It reflects many basic characteristics of human brain functions and is a highly complex nonlinear dynamic learning system.

[0025] <3> A bitmap is an image divided into pixels, each with a number of binary bits representing its color, brightness, and attributes. Bitmaps are typically acquired using devices such as scanners, cameras, video recorders, laser discs, and video disc signal digitizers, which convert analog image signals into digital image data.

[0026] Embodiments of the present invention provide a shared device scheduling method, an apparatus for implementing the shared device scheduling method, and a storage medium storing an executable program for implementing the shared device scheduling method. Regarding the implementation of the shared device scheduling method, the embodiments of the present invention will describe an exemplary implementation process and principles of shared device scheduling.

[0027] like Figure 1 FIG. 1 shows a flow chart of a shared device scheduling method according to an embodiment of the present invention, including steps S11 to S12.

[0028] The shared devices in this embodiment include but are not limited to shared bicycles, shared power-assisted bicycles, shared cars, shared batteries, shared power banks, shared umbrellas, etc. Since there are many types of devices, they will not be listed one by one. However, for the sake of ease of description, the following embodiments mostly use shared bicycles as an example.

[0029] Step S11: Acquire the target area image collected in real time and classify it.

[0030] In some examples, a deep learning neural network model is used to classify the target area image. The specific implementation process is as follows: Figure 2 As shown, the following steps are included:

[0031] Step S21: Acquire multiple historically collected images and perform preliminary scene classification based on image granularity features; wherein the preliminary classification results include point accumulation images, point non-accumulation images, and irrelevant images.

[0032] It should be understood that the dot accumulation image refers to a dot map where the target device has a pile-up phenomenon, the dot non-accumulation image refers to a dot map where the target device does not have a pile-up phenomenon, and the irrelevant image refers to a dot map where no target device appears.

[0033] Furthermore, preliminary scene classification based on image granularity features can be performed manually, and the preliminary classification results can be manually annotated. Alternatively, existing fine-grained image classification algorithms, such as those used for fine-grained image classification, can be used for preliminary scene classification; these fine-grained image classification algorithms identify subclasses. Traditional image classification primarily deals with two categories: semantic-level images and instance-level images. The former, which includes semantic-level image classification tasks such as scene recognition and object recognition, aims to identify objects of different categories; the latter, for example, classifies individuals. Fine-grained image classification lies somewhere in between these two categories. Unlike coarse-grained image classification tasks like object recognition, fine-grained images offer more refined classification accuracy and subtle inter-class differences. Classification often relies solely on small local differences. Within-class differences in fine-grained images are even greater, and they are subject to numerous uncertainties such as pose, lighting, occlusion, and background interference. The classification process of these fine-grained image classification algorithms involves first identifying foreground objects and their local regions, then extracting features from these regions, and processing the resulting features to complete classification.

[0034] Step S22: annotate the historically collected images according to the preliminary scene classification results to form a training set, which is then input into a deep learning neural network model for training to obtain a scene classification model.

[0035] The scene classification model includes at least an input layer, a convolution layer, a pooling layer, a BN layer and an output layer; wherein the convolution layer is used for convolution calculation; the role of the pooling layer is to compress data and data volume, reduce overfitting, mainly by gradually reducing the spatial size of the representation to reduce the number of parameters and calculations in the network, extract important features, and delete redundant noise feature information; the BN layer generally standardizes the features of the input samples, and converts the data into a Gaussian distribution with a mean of 0 and a standard deviation of 1, or a distribution with a range near 0 (if the BN layer is not processed, the sample features will be too dispersed, which will slow down the learning speed or even make it impossible to learn).

[0036] The training process of the scene classification model specifically includes:

[0037] 1) Inputting the point map of the historically collected image into the input layer of the neural network model.

[0038] 2) After using the several convolution layers to perform convolution calculation on the point bitmap, using the pooling layer to compress the convolution calculation result, and then using the batch normalization layer to perform batch normalization processing on the data compression result to obtain an image feature map.

[0039] 3) In the output layer, the image feature map is encoded and extracted using the same number of output neurons as the number of scene categories to obtain a logits vector. The logits vector is then input into a softmax function to obtain the probability that the point map of the historically captured image belongs to each scene category. It should be understood that in deep learning, the logits vector is typically fed into a softmax function; the formula for the softmax function is as follows:

[0040]

[0041] Where Xi is the logits vector output by the i-th output neuron, and Softmax(x)i is the output value of the i-th logits vector after being processed by the softmax function. It should be understood that the softmax function can convert the output values ​​of multiple categories into a probability distribution in the range [0.1].

[0042] 4) Calculate the loss function based on the output of the softmax function and the sample labels of the samples in the training set, and obtain the final scene classification model after supervised learning using the sample labels.

[0043] Preferably, this embodiment provides a loss function with sample bias, which can compensate for the imbalance of sample categories by setting corresponding weights. Generally speaking, samples in deep learning can be distinguished in multiple dimensions. For example, they can be divided into difficult samples and simple samples in terms of difficulty. Difficult samples refer to samples with a large error from the true value label when predicted, and simple samples refer to samples with a small error from the true value label when predicted; for example, according to the long-tail distribution (unbalanced distribution) characteristics of the sample category, the sampling probability of the sample category is different. The traditional loss function does not set a bias for the sample category and cannot cope with the imbalance of the sample category; the loss function in this embodiment sets a sample bias, so the imbalance of the sample category can be compensated by setting corresponding weights, such as setting it to be biased towards difficult samples or biased towards samples with better quality but fewer samples, thereby making the model prediction results more accurate.

[0044] Specifically, the expression of the newly defined loss function in this embodiment is as follows:

[0045] FCB Loss=w i (1-p i ) γ log(p i ); Formula 2)

[0046]

[0047] Among them, Pi represents the probability value of the model prediction result of the current sample belonging to the predicted true label category, N represents the number of all samples in the training set, and n i represents the number of samples with label category i, γ and β are custom configuration parameters, and w i is the sample weight of the label category i. From formula 3), it can be seen that by adjusting the number of samples n i The sample weight w can be adjusted accordingly i .

[0048] Step S23: Inputting the real-time captured target area image into the scene classification model for prediction, thereby obtaining a scene classification result output by the scene classification model. Specifically, the scene classification model constructed based on historically captured images can be used to classify the real-time captured images. The scene classification results output by the model include point accumulation image, point non-accumulation image, and irrelevant image.

[0049] Step S12: For the target area image classified as the point accumulation image, target device counting is performed and whether the target area needs to be scheduled is determined based on the counting result.

[0050] The determination of whether the target area needs to be dispatched based on the counting result is mainly based on whether the technical result exceeds the preset threshold; if it exceeds the preset threshold, it means that dispatching is required, otherwise no dispatching is required. The preset threshold can be set as a fixed value (for example, 50 is set as the preset threshold); it can also be dynamically adjusted according to the size of the target area, for example, a larger preset threshold can be set for a larger area, and conversely, a smaller preset threshold can be set for a smaller area; the preset threshold can also be set according to the traffic importance of the target area, for example, for a traffic hub area, a slight accumulation of vehicles may cause traffic congestion, so a lower preset threshold can be set; conversely, for a non-hub area, a higher preset threshold can be set; in fact, the present invention does not limit the specific setting rules of the preset threshold.

[0051] It should be noted that device accumulation can be caused by similar devices other than the target device. Therefore, after scene classification, it is necessary to count the target devices to prevent false detections and prevent interference from other devices on the road. For example, shared bikes, such as Hellobike, Mobike, and ofo, are all identical, differing only in appearance, making false detections very likely.

[0052] In this embodiment, the process of counting target devices for the target area image classified as the dot accumulation image includes the following steps:

[0053] Step S31: Use the target detection model to perform target detection on the target area image to obtain prediction boxes of all target positions.

[0054] The target detection model can use a two-stage detection algorithm (Two-Stage) or a one-stage detection algorithm (One-Stage). The two-stage detection algorithm (Two-Stage) divides the detection problem into two stages. The first stage generates candidate regions (region proposals), and the second stage classifies the candidate regions. Typical representatives of the two-stage detection algorithm include R-CNN algorithms, such as R-CNN (Region CNN) algorithm, Fast R-CNN (Fast Region CNN) algorithm, Faster R-CNN (Faster Region CNN) algorithm, etc. The one-stage detection algorithm does not need to generate candidate regions, but directly generates the category probability and position coordinate value of the object; typical representatives of the one-stage algorithm include YOLO algorithm and SSD algorithm, etc. Since there are many target detection algorithms, they will not be described here one by one.

[0055] Specifically, the target detection model uses several convolutional layers to extract image features of the target area image and outputs feature maps at multiple depths; the features corresponding to the prediction boxes of corresponding sizes are calculated on the feature maps at different depths to obtain the output results at each depth; a comprehensive calculation is performed based on the output results at each depth to output the coordinate positions, target categories and confidence levels of all prediction boxes corresponding to the target positions on the original target area image.

[0056] It's important to note that a convolutional layer in a convolutional neural network consists of several convolutional units, each optimized through backpropagation. The purpose of a convolution operation is to extract distinct features from the input. While the first convolutional layer may only extract low-level features like edges, lines, and corners, a network with more layers can iteratively extract more complex features from these low-level features.

[0057] In some examples, the optimal prediction box is selected from multiple prediction boxes corresponding to a target position and redundant prediction boxes are removed to obtain a prediction box uniquely corresponding to the target position; the prediction box and the prediction box information include the position coordinates, target category and confidence of the prediction box in the target area image.

[0058] Due to the large number of prediction boxes, a target location may be predicted by multiple prediction boxes. It is necessary to select the optimal prediction box and remove redundant prediction boxes. Therefore, it is necessary to use the non-maximum suppression algorithm to remove some redundant boxes. The specific process includes: first selecting a prediction box with the highest confidence, then deleting prediction boxes of the same category whose overlap with this area exceeds a certain threshold, and repeating the above operation for the remaining prediction boxes until all prediction boxes are traversed. Therefore, after removing duplicate prediction boxes, the unique corresponding location coordinates, target category, and confidence level of each target location are obtained.

[0059] The overlap evaluation is based on the IOU (Intersection Over Union) standard, which is a standard for detecting the accuracy of corresponding objects in a specific dataset. It is calculated by dividing the overlapping area of ​​the two prediction boxes by the sum of the areas of the two prediction boxes. The larger the ratio, the higher the overlap of the two prediction boxes, and vice versa.

[0060] Step S32: converting the target area image into an HSI image, and selecting a color interval for representing the target device based on the HSI image.

[0061] In some examples, the target area image is converted into an HSI image based on the HSI color model. The HSI color model is based on the human visual system and uses H (Hue) to represent hue, S (Saturation) to represent saturation, and I (Intensity) to represent brightness to describe color. Hue H (Hue) refers to the wavelength transmitted or reflected by an object. It is more commonly identified by colors such as red, orange, or green, and is measured as a value from 0 to 360 degrees. Saturation (Saturation), also known as chroma, refers to the intensity or purity of the color; saturation represents the ratio of gray to hue and is measured from 0% (gray) to 100% (fully saturated); brightness (Intensity) refers to the relative lightness or darkness of the color, usually measured as a percentage from 0% (black) to 100% (white).

[0062] Furthermore, the H (Hue) and S (Saturation) components of the HSI image are used to select a color range to represent the target device. Taking shared bicycles as an example, the color range used to represent Hellobikes is blue, the color range used to represent Mobikes is orange, and the color range used to represent ofo bikes is yellow.

[0063] Step S33: determining whether there is a false detection based on whether there is an intersection between the prediction box of the target position and the image mask of the color interval, and eliminating the target devices determined to be falsely detected and counting them.

[0064] The concept of an image mask is borrowed from the PCB manufacturing process. In semiconductor manufacturing, many chip process steps utilize photolithography. The pattern "negative" used in these steps is called a mask. Its function is to cover selected areas on the silicon wafer with an opaque pattern template, so that subsequent etching or diffusion will only affect areas outside the selected area. Similarly, a pattern mask uses a selected image, graphic, or object to block the processed image, thereby controlling the image processing area or processing process.

[0065] Judging whether it is a false detection based on whether there is an intersection area between the prediction box of the target position and the image mask of the color interval means: if there is an intersection between the prediction box of the target position and the image mask of the color interval, it means that the target position is indeed a target device and should be counted; if there is no intersection between the prediction box of the target position and the image mask of the color interval, it means that the target position is not actually a target device and should not be counted, and the prediction box of this target position should be removed.

[0066] Furthermore, for a target area image classified as a non-point accumulation image, it is determined that the target area does not need to be scheduled.

[0067] After deleting all prediction boxes that were falsely detected, the number of prediction boxes remaining at the target location is calculated to be the final number of target devices. If the number of target devices exceeds a certain threshold, it indicates that device accumulation has occurred, meaning that the current image is a point-stacked image, and therefore the target area in the image needs to be scheduled. Conversely, if the number of target devices does not exceed a certain threshold, it indicates that device accumulation has not occurred, meaning that the current image is a point-non-stacked image or an unrelated image, and therefore no target area in the image needs to be scheduled.

[0068] The shared device scheduling method provided in the embodiment of the present invention can be implemented on the terminal side or the server side. As for the hardware structure of the shared device scheduling terminal, please refer to Figure 3 , is an optional hardware structure diagram of the electronic terminal 300 provided in an embodiment of the present invention. The terminal 300 can be a mobile phone, a computer device, a tablet device, a personal digital processing device, a factory background processing device, etc. The electronic terminal 300 includes: at least one processor 301, a memory 302, at least one network interface 304 and a user interface 306. The various components in the device are coupled together through a bus system 305. It can be understood that the bus system 305 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 305 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 3 Various buses are labeled as bus systems.

[0069] The user interface 306 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0070] It will be appreciated that the memory 302 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM) or a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0071] The memory 302 in the embodiment of the present invention is used to store various categories of data to support the operation of the electronic terminal 300. Examples of such data include: any executable program for operating on the electronic terminal 300, such as an operating system 3021 and an application 3022; the operating system 3021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 3022 can include various applications, such as a media player (MediaPlayer), a browser (Browser), etc., for implementing various application services. The shared device scheduling method provided in the embodiment of the present invention can be included in the application 3022.

[0072] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 301. Processor 301 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in processor 301 or by software instructions. The above processor 301 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc. Processor 301 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor 301 may be a microprocessor or any conventional processor. The steps of the shared device scheduling method provided in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium located in a memory. The processor reads information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0073] In an exemplary embodiment, the electronic terminal 300 may be configured to execute the aforementioned method using one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs).

[0074] like Figure 4 , which shows a schematic diagram of the structure of a shared device scheduling apparatus according to an embodiment of the present invention. The shared device scheduling apparatus 400 in this embodiment includes a scene classification module 401 and a scheduling allocation module 402.

[0075] The scene classification module 401 is used to obtain the target area image collected in real time and classify it.

[0076] In some examples, the scene classification module 401 uses a deep learning neural network model to classify the target area image.

[0077] Furthermore, the process of classifying the target area image by the deep learning neural network model includes: obtaining multiple historically collected images and performing preliminary scene classification based on image granularity features; annotating the historically collected images according to the preliminary scene classification results to form a training set, which is input into the deep learning neural network model and then trained to obtain a scene classification model; inputting the target area image collected in real time into the scene classification model for prediction to obtain the scene classification result output by the scene classification model.

[0078] In some examples, the deep learning neural network model includes an input layer, several convolutional layers, a pooling layer, and a batch normalization layer output layer; the training process of the deep learning neural network model includes: inputting the point map of the historically collected image into the input layer; using the several convolutional layers to perform convolution calculations on the point map, using the pooling layer to compress the convolution calculation results, and then using the batch normalization layer to perform batch normalization on the data compression results to obtain an image feature map; in the output layer, using the same number of output neurons as the number of scene categories to encode and extract the image feature map to obtain a logits vector, and after inputting the logits vector into the softmax function, the probability that the point map of the historically collected image belongs to each scene category is obtained.

[0079] Furthermore, the deep learning neural network model uses a sample-biased loss function during training to compensate for the imbalance of sample categories by setting corresponding weights.

[0080] The scheduling allocation module 402 is used to count target devices for a target area image classified as a point accumulation image and determine whether the target area needs to be scheduled based on the counting result.

[0081] In some examples, the process of counting target devices on a target area image of a classified point accumulation image includes: performing target detection on the target area image using a target detection model to obtain prediction boxes for all target positions; converting the target area image into an HSI image, and selecting a color interval to characterize the target device based on the HSI image; determining whether there is a false detection based on whether there is an intersection area between the prediction box of the target position and the image mask of the color interval, and counting the target devices determined to be falsely detected after eliminating them.

[0082] Furthermore, target detection is performed on the target area image using a target detection model to obtain a prediction box of the target position, including: selecting the optimal prediction box from multiple prediction boxes corresponding to a target position and removing redundant prediction boxes to obtain a prediction box uniquely corresponding to each target position; the prediction box information of the prediction box includes the position coordinates, target category and confidence of the prediction box in the target area image.

[0083] It should be noted that the shared device scheduling system provided in this embodiment is similar to the shared device scheduling method in the above embodiment, so it will not be described in detail.

[0084] It should be noted that the shared device scheduling system provided in the above embodiment only uses the division of the above-mentioned program modules as an example for explanation of scheduling. In actual applications, the above-mentioned processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the above-mentioned processing. In addition, the image classification device provided in the above embodiment and the image classification method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0085] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the shared device scheduling method when executed by a processor.

[0086] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with a computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0087] In the embodiments provided by the present invention, the computer readable and writable storage medium may include a read-only memory, a random access memory, an EEPROM, a CD-ROM or other optical disk storage device, a magnetic disk storage device or other magnetic storage device, a flash memory, a USB flash drive, a mobile hard disk, or any other medium that can be used to store desired program code in the form of instructions or data structures and can be accessed by a computer. In addition, any connection can be appropriately referred to as a computer-readable medium. For example, if the instruction is sent from a website, server or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL) or wireless technologies such as infrared, radio and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL or wireless technologies such as infrared, radio and microwaves are included in the definition of the medium. However, it should be understood that computer readable and writable storage media and data storage media do not include connections, carriers, signals or other temporary media, but are intended to be non-temporary, tangible storage media. Disk and disc, as used in this application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

[0088] In summary, the present invention provides a shared equipment scheduling method, device, terminal and storage medium. The present invention realizes the automatic identification of shared equipment scheduling. By classifying real-time images and counting specific devices in the images, the real-time analysis of images is completed accurately, in real time and efficiently, which makes up for the lack of ability of offline staff in self-building scheduling tasks, saves the manpower cost of manual approval, and avoids the cost loss to the company caused by the complexity of offline personnel composition. It is an important technical solution in the field of risk control in the sharing industry.

[0089] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A shared device scheduling method, characterized in that: include: Acquire the target area image collected in real time and classify it; For target area images classified as point accumulation images, target device counts are performed and whether the target area needs to be dispatched is determined based on the count results; The target device counting is performed on the target area image of the classified point accumulation image, including: Performing target detection on the target area image using a target detection model to obtain a predicted frame of the target position; Converting the target area image into an HSI image, and selecting a color interval for representing the target device based on the HSI image; Whether there is a false detection is determined based on whether there is an intersection area between the prediction box of the target position and the image mask of the color interval, and the target devices determined to be falsely detected are eliminated and then counted.

2. The shared device scheduling method according to claim 1, wherein: It also includes using a deep learning neural network model to classify the target area image.

3. The shared device scheduling method according to claim 2, wherein: The process of classifying the target area image by the deep learning neural network model includes: Acquire multiple historically collected images and perform preliminary scene classification based on image granularity features; The historically collected images are annotated according to the preliminary scene classification results to form a training set, which is then input into a deep learning neural network model for training to obtain a scene classification model; The target area image is input into the scene classification model for prediction to obtain a scene classification result output by the scene classification model.

4. The shared device scheduling method according to claim 3, wherein: The deep learning neural network model includes an input layer, a convolutional layer, a pooling layer, a batch normalization layer, and an output layer; the training process of the deep learning neural network model includes: Inputting the point map of the historically collected image into the input layer; After performing convolution calculation on the point bitmap using the convolution layer, compressing the convolution calculation result using the pooling layer, and then performing batch normalization processing on the data compression result using the batch normalization layer to obtain an image feature map; In the output layer, the image feature map is encoded and extracted using the same number of output neurons as the number of scene categories to obtain a logits vector, and the logits vector is input into the softmax function to obtain the probability that the point map of the historically collected image belongs to each scene category.

5. The shared device scheduling method according to claim 3, wherein: The deep learning neural network model uses a sample-biased loss function during training to compensate for the imbalance of sample categories by setting corresponding weights.

6. The shared device scheduling method according to claim 1, wherein: Performing target detection on the target area image using a target detection model to obtain a prediction frame of the target position, including: selecting an optimal prediction frame from multiple prediction frames corresponding to a target position and removing redundant prediction frames to obtain a prediction frame uniquely corresponding to each target position.

7. A shared device scheduling device, characterized in that: include: The scene classification module is used to obtain and classify the target area images collected in real time; A scheduling allocation module is used to count target devices in a target area image classified as a point accumulation image and determine whether the target area needs to be scheduled based on the counting result; The target device counting is performed on the target area image of the classified point accumulation image, including: Performing target detection on the target area image using a target detection model to obtain a predicted frame of the target position; Converting the target area image into an HSI image, and selecting a color interval for representing the target device based on the HSI image; Whether there is a false detection is determined based on whether there is an intersection area between the prediction box of the target position and the image mask of the color interval, and the target devices determined to be falsely detected are eliminated and then counted.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the shared device scheduling method according to any one of claims 1 to 6 is implemented.

9. An electronic terminal, characterized in that: include: processor and memory; The memory is used to store computer programs; The processor is configured to execute the computer program stored in the memory, so as to enable the terminal to execute the shared device scheduling method according to any one of claims 1 to 6.

Images