Image analysis method, system, device and storage medium based on big data technology

Abstract

The present invention belongs to the field of data processing technology, and specifically relates to an image analysis method, system, device and storage medium based on big data technology. The method includes presetting multiple recognition parameter information, using the multiple recognition parameter information in sequence to perform a first recognition process on a learning image, and obtaining recognition results corresponding to different recognition parameter information; according to the recognition result and the target area of ​​the learning image stored in advance, obtaining the recognition threshold and the number of recognitions corresponding to each recognition parameter information, and selecting the second recognition parameter information for the second recognition process based on the recognition threshold and the number of recognitions corresponding to each recognition parameter information; obtaining an image to be recognized, and performing a second recognition process on the image to be recognized based on the second recognition parameter information to obtain a second recognition result. The present invention can improve the efficiency of the recognition model in recognizing a target object from any image while ensuring the recognition accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to an image analysis method, system, device and storage medium based on big data technology. Background Art

[0002] In today's society, image data is ubiquitous. They are widely used in many fields such as medical treatment, transportation, security monitoring, industrial automation, smart home and multimedia processing. The application of big data computing for image recognition can be applied to various fields. A similar prior art Chinese patent application document with publication number CN112131424A provides a distributed image analysis method and system, including: obtaining multiple candidate images and classifying and marking them; determining an index image in multiple candidate images and extracting features to obtain feature values; comparing the similarity of the feature values ​​of the index image with the remaining images in the database and sorting and outputting them; determining similar images according to the output results, and decoding the similar images to generate image files. It can select index images one by one from multiple candidate image sets in the database, extract features from the index images, and compare and sort them with two adjacent candidate images respectively, so as to achieve the purpose of distributed image analysis. A similar prior art also has a public document with publication number CN106020436A, which discloses an image analysis device and an image analysis method. The method includes an image analysis device according to one aspect, including a display, a camera, an estimation unit, a creation unit, a guidance unit and a display control unit. The camera captures an image of an object located in front of a display. The estimation unit estimates the shape and position of the object whose image has been captured by the camera. The creation unit creates a display area on the display according to the shape and position of the object. The guidance unit calculates the guidance direction of the object according to the shape and position of the object. The display control unit displays guidance instructions based on the guidance direction on the display area. However, neither of the above two documents considers improving the efficiency of the recognition model in identifying the target object from any image while ensuring the recognition accuracy. Therefore, the present invention provides an image analysis method, system, device and storage medium based on big data technology. Summary of the invention

[0003] When identifying the target object, in order to eliminate recognition omissions, the recognition area is moved with a smaller sliding size in the recognition image, which will lead to an increase in the number of recognition areas. Since the recognition processing time increases with the increase in the number of recognition areas, in order to improve the processing efficiency, the number of recognitions is reduced by increasing the sliding size of the recognition area. However, as the sliding size of the recognition area increases, the target object may be ignored and the probability of recognition omissions will increase. It is found that the setting of the sliding size of the recognition area is data that affects the recognition processing efficiency and recognition accuracy. Therefore, it is necessary to obtain a suitable sliding size of the recognition area to improve the recognition efficiency while ensuring the recognition accuracy.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following image analysis method based on big data technology, which is implemented by executing the following steps:

[0005] Pre-setting a plurality of recognition parameter information, the recognition parameter information including a recognition area size, a recognition area sliding size and a first threshold value, different recognition parameter information having different recognition area sliding sizes and first threshold values, sequentially using the plurality of recognition parameter information to perform a first recognition process on a learning image, the learning image being an image including a target object, pre-storing an object area of ​​the target object in the learning image, obtaining recognition results corresponding to different recognition parameter information, the recognition results including accuracy and a target area, the target area being an area where the target object is located;

[0006] According to the recognition result and the object area of ​​the learning image stored in advance, obtaining a recognition threshold and a recognition number corresponding to each of the recognition parameter information, and obtaining second recognition parameter information for second recognition processing based on the recognition threshold and the recognition number corresponding to each of the recognition parameter information;

[0007] Acquire an image to be identified, perform a second identification process on the image to be identified based on the second identification parameter information to obtain a second identification result, wherein the second identification result includes an identification area where a target object in the image to be identified is located and a corresponding accuracy, and save the second identification result.

[0008] As a preferred technical solution of the present invention, performing a first recognition process on the learning image includes the following steps:

[0009] Obtain one piece of recognition parameter information from all the recognition parameter information, input the recognition parameter information and the learning image into a recognition model, set a recognition area size, the recognition model obtains all recognition areas of the learning image according to the recognition area size and the recognition area sliding size of the recognition parameter information, the recognition model identifies all the recognition areas and obtains a recognition result, the recognition result includes a target area and an accuracy of the target object in the recognition area, calculates an overlap rate between the object area and the target area according to the target area and an object area of ​​the target object in the learning image stored in advance, compares the overlap rate with a first threshold corresponding to the first recognition parameter information, if the overlap rate is greater than or equal to the first threshold, determines that the recognition area contains the target object, if the overlap rate is less than the first threshold, determines that the recognition area does not contain the target object, obtains a total number of all recognition areas determined to contain the target object, and also obtains the accuracy of the corresponding recognition area, uses the total number as the number of recognitions of the first recognition parameter information, obtains a minimum value in the accuracy, uses the minimum value as the recognition threshold of the first recognition parameter information, and associates and saves the recognition threshold, the number of recognitions, and the first recognition parameter information;

[0010] Repeat the above steps to obtain the recognition times and corresponding recognition thresholds of all the recognition parameter information.

[0011] As a preferred technical solution of the present invention, obtaining second identification parameter information for second identification processing includes the following steps:

[0012] Obtain the identification threshold and the number of identification times for each identification parameter information, add the number of identification times corresponding to all the identification parameter information to obtain a first total number, divide the first total number by a specific value to obtain an average number, the specific value is equal to the level number of the identification parameter information plus one, the level number of the identification parameter information is pre-defined when setting the identification parameter information, determine the best identification parameter information based on the average number, and use the best identification parameter information as the second identification parameter information.

[0013] As a preferred technical solution of the present invention, performing a second recognition process on the image to be recognized based on the second recognition parameter information comprises the following steps:

[0014] First, the first second recognition parameter information among all the second recognition parameter information is obtained, and the recognition area, the sliding size of the recognition area and the recognition threshold corresponding to the second recognition parameter information are also obtained, and the recognition area of ​​the image to be recognized is obtained according to the recognition area size and the sliding size of the recognition area, and the accuracy of each recognition area is obtained by using the recognition model, and the accuracy is compared with the recognition threshold of the second recognition parameter information. If the accuracy is greater than or equal to the recognition threshold, the corresponding recognition area is used as the next area to be recognized of the second recognition parameter information, and if the accuracy is less than the recognition threshold, the recognition area is excluded;

[0015] Acquire the next second identification parameter information, and repeat the previous step until all the second identification parameter information is completed.

[0016] As a preferred technical solution of the present invention, obtaining all recognition areas of the learning image according to the recognition area size and the recognition area sliding size of the first recognition parameter information includes the following steps:

[0017] Take the upper left corner of the learning image as the upper left corner of the recognition area to obtain a rectangular area with a width k and a height h of the recognition area size, take the rectangular area as the recognition area, and after the recognition model completes the recognition of the recognition area, obtain the sliding width and sliding height of the sliding size of the recognition area, move the rectangular area horizontally to the right by the distance of the sliding width, and obtain a new rectangular area each time it moves, and take the new rectangular area as the recognition area. After each horizontal movement to the right, determine whether the right boundary of the rectangular area coincides with the right boundary of the learning image. If not, continue to move horizontally to the right by the distance of the sliding width until the rectangular area The right boundary of the rectangular area coincides with the right boundary of the learning image, and then the rectangular area is moved downward in the vertical direction by the distance of the sliding height to obtain a new rectangular area, and the new rectangular area is identified as the recognition area, and then the rectangular area is moved leftward in the horizontal direction by the distance of the sliding width until the left boundary of the rectangular area coincides with the left boundary of the learning image, and then it is determined whether the lower boundary of the rectangular area coincides with the lower boundary of the learning image. If not, continue to move the rectangular area downward in the vertical direction by the distance of the sliding height, and repeat this step until the recognition processing is completed for all areas of the learning image. If they coincide, the processing is terminated.

[0018] As a preferred technical solution of the present invention, calculating the overlap rate between the object area and the target area includes the following steps:

[0019] The overlap rate R is calculated using a first formula, which is: R=S3 / (S1+S2-S3), wherein S1 refers to the area of ​​the object region, S2 refers to the area of ​​the target region, and S3 refers to the area of ​​the overlapping region between the object region and the target region.

[0020] As a preferred technical solution of the present invention, after the recognition model is used to recognize the recognition area and the accuracy of each recognition area is obtained, the following steps are performed:

[0021] Obtaining the accuracy of each of the recognition areas, determining whether there is a recognition area whose accuracy is greater than or equal to the recognition threshold, and if so, marking the recognition area whose accuracy is greater than or equal to the recognition threshold. After determining all the recognition areas, selecting one of the marked recognition areas, setting a moving distance, moving the recognition area by the moving distance in each movable direction to obtain a new recognition area, identifying the new recognition area with the recognition model, obtaining the corresponding accuracy and object area, obtaining the accuracy of the new recognition area in each movable direction, and if the accuracy is greater than or equal to the recognition threshold, continuing to move the recognition area along the movable direction, performing recognition processing on the moved recognition area to obtain a corresponding recognition result, until the recognition area moves to the boundary of the image to be recognized, and terminating the processing;

[0022] Determine whether there are other marked identification areas, and if so, repeat the previous step until all marked identification areas are identified.

[0023] The present invention also provides an image analysis method system based on big data technology, comprising the following modules:

[0024] A first recognition processing module is used to pre-set multiple recognition parameter information, the recognition parameter information includes a recognition area size, a recognition area sliding size and a first threshold, different recognition parameter information has different recognition area sliding sizes and corresponding first thresholds, and sequentially use the multiple recognition parameter information to perform first recognition processing on a learning image, the learning image refers to an image including a target object, pre-store an object area of ​​the target object in the learning image, and obtain recognition results corresponding to different recognition parameter information, the recognition results include accuracy and a target area, and the target area refers to an area where the target object is located;

[0025] a recognition parameter selection module, configured to obtain a recognition threshold and a recognition number corresponding to each of the recognition parameter information according to the recognition result and the object area of ​​the extracted and stored learning image, and to obtain second recognition parameter information for second recognition processing based on the recognition threshold and the recognition number corresponding to each of the recognition parameter information;

[0026] The second recognition processing module is used to perform a second recognition processing on the image to be recognized based on the second recognition parameter information to obtain a second recognition result, wherein the second recognition result includes a recognition area where the target object in the image to be recognized is located and a corresponding accuracy, and the second recognition result is saved.

[0027] The present invention also provides an electronic device, characterized in that it comprises: a processor, a memory and a communication bus, wherein the processor and the memory communicate with each other via the communication bus;

[0028] The memory is used to store computer programs;

[0029] The processor is used to execute the program stored in the memory to implement any of the above-mentioned image analysis methods based on big data technology.

[0030] The present invention also provides a storage medium, which stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute any one of the above-mentioned image analysis methods based on big data technology.

[0031] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0032] In the present invention, multiple pieces of recognition parameter information are pre-set, and the multiple pieces of recognition parameter information are used in sequence to perform a first recognition process on the learning image, and the recognition results corresponding to different recognition parameter information are obtained; according to the recognition result and the target area of ​​the learning image stored in advance, the recognition threshold and the number of recognitions corresponding to each recognition parameter information are obtained, and the second recognition parameter information for the second recognition process is selected based on the recognition threshold and the number of recognitions corresponding to each recognition parameter information; the image to be recognized is obtained, and the second recognition process is performed on the image to be recognized based on the second recognition parameter information to obtain the second recognition result. The present invention can improve the efficiency of the recognition model in recognizing the target object from any image while ensuring the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flowchart of the steps of the image analysis method based on big data technology of the present invention;

[0034] Figure 2 An example diagram of identification parameter information having three pieces of identification parameter information of the present invention;

[0035] Figure 3 A schematic diagram of a learning image or an image to be recognized according to the present invention;

[0036] Figure 4 A schematic diagram of determining optimal recognition parameter information of the present invention;

[0037] Figure 5 This is a structural diagram of the image analysis system based on big data technology of the present invention;

[0038] In the figure: 301, recognition area; 302, target object; 303, rectangular area occupied by the target object. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0041] The present invention provides Figure 1 The image analysis method based on big data technology is implemented by executing the following steps:

[0042] Step S1, pre-set multiple recognition parameter information, the recognition parameter information includes recognition area size, recognition area sliding size and a first threshold, different recognition parameter information has different recognition area sliding size and first threshold, use multiple recognition parameter information in sequence to perform first recognition processing on the learning image, the learning image refers to an image including a target object, pre-store the object area of ​​the target object in the learning image, obtain recognition results corresponding to different recognition parameter information, the recognition result includes accuracy and target area, the target area refers to the area where the target object is located.

[0043] Specifically, in order to improve the efficiency of the recognition model in identifying the target object from any image under the premise of ensuring the recognition parameter information, multiple recognition parameter information is preset, and the recognition parameter information includes the recognition area size, the recognition area sliding size and the first threshold. The recognition area size is used to set the size of the recognition area extracted from the image, such as Figure 2 The following is an example diagram of identification parameter information having three pieces of identification parameter information, such as Figure 3 FIG. 3 is a schematic diagram of a learning image or an image to be recognized, wherein 301 is a recognition area, and the sliding size of the recognition area includes a sliding width and a sliding height, such as Figure 3, respectively, are the width and height of the recognition area 301. The first threshold is a threshold used to determine the percentage of the actual target object in each recognition area in the subsequent first recognition process. For example, if the first threshold is 0.6, the target object area occupies more than 60% of the total area of ​​the recognition area before determining that the target object exists in the recognition area. The object area of ​​the target object in the learning image is pre-stored. The object area refers to the rectangular area occupied by the target object in the learning image, such as Figure 3 As shown in the figure, 302 is the target object, 303 is the rectangular area occupied by the target object, the target object refers to the object to be identified, which can be a person, a car or any other object that needs to be detected. By setting different recognition parameter information, the first recognition processing is performed on the learning image to obtain the recognition results corresponding to the different recognition parameter information. The recognition result includes accuracy and target area. The target area refers to the area where the target object is located, and the accuracy refers to the probability that the target object is included in the recognition area.

[0044] Step S2, according to the recognition results and the object area of ​​the learning image stored in advance, obtain the recognition threshold and recognition times corresponding to each recognition parameter information, and obtain the second recognition parameter information used for the second recognition processing based on the recognition threshold and recognition times corresponding to each recognition parameter information.

[0045] Specifically, in order to obtain the recognition parameter information to be used in the second recognition processing, according to the recognition result obtained by the first recognition processing, the recognition threshold and the number of recognition times corresponding to each recognition parameter information are counted, and the second recognition parameter information to be used in the second recognition processing is obtained based on the recognition parameter information and the recognition threshold and the number of recognition times corresponding to the recognition parameter information. There may be multiple second recognition parameter information. Only by performing the second recognition processing on the image to be recognized through the second recognition parameter information, the recognition area of ​​the second recognition processing can be reduced, the number of recognitions can be effectively reduced, and the recognition efficiency can be improved. The second recognition result is also obtained according to the obtained recognition threshold to ensure the accuracy of recognition.

[0046] Step S3, obtaining an image to be identified, performing a second identification process on the image to be identified based on the second identification parameter information to obtain a second identification result, the second identification result including an identification area where the target object in the image to be identified is located and a corresponding accuracy, and saving the second identification result.

[0047] Specifically, an image to be identified is obtained, and a second identification process is performed on the image to be identified based on second identification parameter information. The specific second identification process method will be explained in detail later. After the second identification result is obtained, the corresponding second identification result is saved.

[0048] Through the above method, the first recognition processing module first performs recognition processing on the learning image according to the set multiple recognition parameter information to obtain the recognition result, and then selects the second recognition parameter information for the second recognition processing according to the recognition result, so as to reduce the time of the second recognition processing while ensuring the recognition accuracy, and improve the efficiency of identifying the target object from the image.

[0049] Furthermore, performing a first recognition process on the learning image comprises the following steps:

[0050] Obtain one piece of recognition parameter information from all recognition parameter information, input the recognition parameter information and the learning image into the recognition model, set the recognition area size, the recognition model obtains all recognition areas of the learning image according to the recognition area size and the recognition area sliding size of the recognition parameter information, the recognition model identifies all recognition areas and obtains recognition results, the recognition results include the target area and accuracy of the target object in the recognition area, calculate the overlap rate of the object area and the target area according to the target area and the object area of ​​the target object in the learning image stored in advance, compare the overlap rate with the first threshold corresponding to the first recognition parameter information, if the overlap rate is greater than or equal to the first threshold, determine that the recognition area contains the target object, if the overlap rate is less than the first threshold, determine that the recognition area does not contain the target object, obtain the total number of all recognition areas determined to contain the target object, and also obtain the accuracy of the corresponding recognition area, use the total number as the number of recognitions of the first recognition parameter information, obtain the minimum value in the accuracy, use the minimum value as the recognition threshold of the first recognition parameter information, and associate the recognition threshold, the number of recognitions and the first recognition parameter information for storage;

[0051] Repeat the above steps to obtain the recognition times and corresponding recognition thresholds of all recognition parameter information.

[0052] Specifically, the first identification parameter information in the identification parameter information is obtained, such as obtaining Figure 2The first recognition parameter information in the image is obtained according to the recognition area size and the corresponding sliding width and sliding height, and all recognition areas of the learning image are obtained. The recognition model is used to identify all recognition areas to obtain the recognition result. The recognition result includes the target area and accuracy of the target object in the recognition area. Since the object area of ​​the target object in the learning image is stored in advance, the object area is the area where the target object is located. The overlap rate of the object area and the target area is calculated. By comparing the overlap rate and the first threshold, it is determined whether the recognition area contains the target object. If the overlap rate is greater than or equal to the first threshold, it means that the proportion of the target object in the recognition area is greater than the first threshold, and it is determined that the recognition area contains the target object. If the overlap rate is less than the first threshold, it means that the proportion of the target object in the recognition area is relatively small, and it is determined that the recognition area does not contain the target object. Since multiple recognition areas are detected, the total number of all recognition areas determined to contain the target object is obtained, and the corresponding accuracy is also obtained. The accuracy of all accuracies is obtained. Minimum accuracy, taking the total number as the number of recognitions of the first recognition parameter information, taking the minimum accuracy as the recognition threshold of the first recognition parameter information, associating and saving the first recognition parameter information and the corresponding number of recognitions and the corresponding recognition threshold, judging whether there is any recognition parameter information that has not been used, if so, using the next recognition parameter information until all recognition parameter information is used, usually the sliding width and sliding height of the first recognition parameter information are set relatively large, and the sliding width and sliding height of the subsequent recognition parameter information are set relatively small, so that more recognition areas can be obtained when the first recognition processing is performed on the learning image, thereby eliminating recognition omissions, because, for example, if a recognition area that divides the target object into two parts is set in the image, it may not be certain that the target object exists in the recognition area, so by setting multiple different recognition parameter information to move the recognition area in a smaller range on the entire learning image, the recognition area is recognized, recognition omissions are eliminated, and recognition accuracy is improved.

[0053] Further, obtaining second recognition parameter information and a second recognition threshold for second recognition processing includes the following steps:

[0054] Obtain the identification threshold and identification times for each identification parameter information, add the identification times corresponding to all the identification parameter information to obtain a first total number, divide the first total number by a specific value to obtain an average number, the specific value is equal to the level number of the identification parameter information plus one, the level number of the identification parameter information is pre-defined when setting the identification parameter information, determine the best identification parameter information based on the average number, and use the best identification parameter information as the second identification parameter information.

[0055] Specifically, in order to obtain the second identification parameter information for the second identification processing, the identification threshold and the number of detections of each identification parameter information are obtained, the number of identifications corresponding to all the identification parameter information is added to obtain the first total number of times, and the first total number of times is divided by a specific value to obtain the average number of times, and the specific value is equal to the number of levels of the identification parameter information plus one, and the number of levels of the identification parameter information is predefined when setting the identification parameter information, such as Figure 2 As shown, the levels of the recognition parameter information are 1, 2, and 3 respectively. After calculating and dividing the average number of times, the best recognition parameter information is determined according to the average number of times. Assuming that the recognition number of times corresponding to the recognition parameter information 1 is 10, the recognition number of times corresponding to the recognition parameter information 2 is 8, and the recognition number of times corresponding to the recognition parameter information 3 is 4, the recognition number of times corresponding to the remaining assumed recognition parameter information is 8 in total, and a specific value is set. Assuming that the specific value is set to 3, the first total number is 30, and the average number of recognitions is 30 / 3=10. Figure 4 The diagram shows the optimal identification parameter information. Figure 4 Where p1 is the recognition parameter information of level 1, p2 is the recognition parameter information of level 2, p3 is the recognition parameter information of level 3, and p4 represents all the remaining recognition parameter information. The total number of recognition times is divided into three parts, and the position where the total number of recognition times is evenly divided is taken as the best position. The recognition parameter corresponding to the best position is determined as the second recognition parameter information, such as Figure 4 The best recognition parameters shown are p1 and p3, which are subsequently used as second recognition parameter information for second recognition processing. The second recognition parameter information used for the second recognition processing is screened out from more recognition parameter information through the above steps, thereby reducing the time of the second recognition processing and improving the recognition efficiency.

[0056] Further, performing a second recognition process on the image to be recognized based on the second recognition parameter information includes the following steps:

[0057] First, the first second recognition parameter information among all the second recognition parameter information is obtained, and the recognition area, the sliding size of the recognition area and the recognition threshold corresponding to the second recognition parameter information are also obtained, and the recognition area of ​​the image to be recognized is obtained according to the recognition area size and the sliding size of the recognition area, and the accuracy of each recognition area is obtained by using the recognition model, and the accuracy and the recognition threshold of the second recognition parameter information are compared. If the accuracy is greater than or equal to the recognition threshold, the corresponding recognition area is used as the next area to be recognized of the second recognition parameter information, and if the accuracy is less than the recognition threshold, the recognition area is excluded;

[0058] The next second identification parameter information is obtained, and the previous step is repeated until all the second identification parameter information is completed.

[0059] Specifically, after the first recognition processing of the learning image, the second recognition parameter information used for the second recognition processing is screened out, the first second recognition parameter information among all the second recognition parameter information is first obtained, and the corresponding recognition area size, recognition area sliding size and recognition threshold are obtained. The recognition area of ​​the image to be recognized is obtained according to the area size and the recognition area sliding size. At this time, the recognized area is the entire range of the image to be recognized. The specific method for obtaining the recognition area is the same as the method for obtaining the recognition area of ​​the learning image, which will be explained in detail later. The recognition model is used to identify all the recognition areas to obtain the corresponding accuracy, and the accuracy and the recognition threshold of the second recognition parameter information are compared. If the accuracy If the accuracy is greater than or equal to the recognition threshold, the recognition area corresponding to the accuracy is used as the area to be recognized by the next second recognition parameter information for recognition processing; if the accuracy is less than the recognition threshold, the corresponding recognition area is not recognized when the next second recognition parameter is used for recognition processing. After all recognition areas are recognized, the next second recognition parameter is obtained, and the previously obtained area to be recognized is used as the recognition area for this recognition. Since some areas of the image to be recognized were excluded in the last recognition processing, the current recognition is not all areas of the image to be processed, thereby reducing the recognition processing time and improving the recognition efficiency. The above process is repeated until the second recognition processing of all second recognition parameter information is completed.

[0060] Further, all recognition areas of the learning image are obtained according to the recognition area size and the recognition area sliding size of the first recognition parameter information, including the following steps:

[0061] Take the upper left corner of the learning image as the upper left corner of the recognition area to obtain a rectangular area with a width k and a height h of the recognition area size. Take the rectangular area as the recognition area. After the recognition model completes the recognition of the recognition area, obtain the sliding width and sliding height of the sliding size of the recognition area. Move the rectangular area horizontally to the right by the sliding width. Each time it moves, a new rectangular area is obtained. The new rectangular area is also used as the recognition area. After each horizontal movement to the right, determine whether the right boundary of the rectangular area coincides with the right boundary of the learning image. If not, continue to move horizontally to the right by the sliding width until the rectangular area coincides with the right boundary of the learning image. The right boundary of the area coincides with the right boundary of the learning image, and then the rectangular area is moved downward in the vertical direction by the sliding height to obtain a new rectangular area, and the new rectangular area is identified as the recognition area, and then the rectangular area is moved to the left in the horizontal direction by the sliding width until the left boundary of the rectangular area coincides with the left boundary of the learning image, and then it is determined whether the lower boundary of the rectangular area coincides with the lower boundary of the learning image. If not, continue to move the rectangular area downward in the vertical direction by the sliding height, and repeat this step until the recognition processing is completed for all areas of the learning image. If they coincide, the processing is terminated.

[0062] Specifically, in order to obtain all the recognition areas of the learning image, a rectangular area is determined with the width k and height h of the recognition area size, such as Figure 4 The figure shows the moving method of the recognition area. The rectangular area is moved as follows: Figure 4 Move in the direction of the arrow shown, slide the rectangular area according to the recognition area sliding size corresponding to the recognition parameter information, first move the sliding width to the right in the horizontal direction to obtain a new rectangular area, use the recognition model to identify the rectangular area, and then continue to move the sliding width to the right until the right boundary of the rectangular area coincides with the right boundary of the learning image, and then move the rectangular area downward in the vertical direction by the sliding height to obtain a new rectangular area, and identify the new rectangular area as the recognition area, and then continue to move the rectangular area to the left in the horizontal direction by the sliding width until the left boundary of the rectangular area coincides with the left boundary of the learning image, and then determine whether the lower boundary of the rectangular area coincides with the lower boundary of the learning image. If not, continue to move the rectangular area downward in the vertical direction by the sliding height, and repeat this step until the recognition processing of all areas of the learning image is completed. If they coincide, end the processing. The above method can be used to identify all recognition areas of the learning image without omission according to the recognition area size and the recognition area sliding size.

[0063] Furthermore, calculating the overlap rate between the object area and the target area includes the following steps:

[0064] The overlap ratio R is calculated using the first formula: R=S3 / (S1+S2-S3), where S1 refers to the area of ​​the object region, S2 refers to the area of ​​the target region, and S3 refers to the area of ​​the overlapping region between the object region and the target region.

[0065] Specifically, the overlap rate between the object area and the target area is calculated by the first formula, and by comparing the overlap rate with the first threshold, it is determined whether the target object exists in the recognition area.

[0066] Furthermore, after the recognition model is used to identify the recognition area and the accuracy of each recognition area is obtained, the following steps are performed:

[0067] Obtain the accuracy of each recognition area, determine whether there is a recognition area with an accuracy greater than or equal to the recognition threshold, if so, mark the recognition area with an accuracy greater than or equal to the recognition threshold, after determining all the recognition areas, select one of the marked recognition areas, set the moving distance, move the recognition area by the moving distance in each movable direction to obtain a new recognition area, use the recognition model to identify the new recognition area, obtain the corresponding accuracy and object area, obtain the accuracy of the new recognition area in each movable direction, if the accuracy is greater than or equal to the recognition threshold, continue to move the recognition area along the movable direction, perform recognition processing on the moved recognition area to obtain the corresponding recognition result, until it moves to the boundary of the image to be recognized, and end the processing;

[0068] Determine whether there are other marked recognition areas. If so, repeat the previous step until all marked recognition areas are recognized.

[0069] Specifically, after all the recognition areas in the image to be recognized are obtained through the first and second recognition parameter information, and all the recognition areas are recognized, the accuracy corresponding to each recognition area is obtained, and it is determined whether there is a recognition area with a corresponding accuracy greater than the recognition threshold among all the recognition areas. If so, the recognition area with an accuracy greater than the recognition threshold is marked. There may be multiple marked recognition thresholds, and one of the recognition areas is selected, and a moving distance d is set. The recognition area is moved from each movable direction by a distance of d to a new recognition area, where the movable direction refers to each direction of the current recognition area, which can be up, down, left or right. However, it should be noted that if the recognition area is moved a distance d in a certain direction, If the area exceeds the boundary of the current image to be identified, the current moving direction is an immovable direction. The accuracy of each movable direction is obtained, and the accuracy and recognition threshold are compared. If the accuracy is greater than the recognition threshold, it means that there is a high possibility that the target object exists in the current direction. The recognition area is subsequently moved along the current direction to obtain the corresponding recognition result. If the accuracy is less than the recognition threshold, it means that there is a low possibility that the target object exists in the current direction. The recognition process is not subsequently performed along this direction. The above method is used to reduce the area that needs to be identified, reduce the time for identification, and improve the efficiency of identification. Then, it is determined whether there are any marked identification areas that have not been processed. If so, the previous step is repeated until all marked identification areas are processed.

[0070] According to another aspect of the embodiment of the present invention, Figure 5 In addition, an image analysis system based on big data technology is provided, including a first recognition processing module, a recognition parameter information selection module, and a second recognition processing module, which are used to implement the image analysis method based on big data technology as described above. The specific functions of each module are as follows:

[0071] A first recognition processing module is used to pre-set multiple recognition parameter information, the recognition parameter information includes a recognition area size, a recognition area sliding size and a first threshold value, different recognition parameter information has different recognition area sliding sizes and corresponding first threshold values, and sequentially use the multiple recognition parameter information to perform first recognition processing on the learning image, the learning image refers to an image including a target object, and the object area of ​​the target object in the learning image is pre-stored, and the recognition results corresponding to different recognition parameter information are obtained, the recognition results include accuracy and target area, and the target area refers to the area where the target object is located;

[0072] A recognition parameter selection module, used to obtain a recognition threshold and a recognition number corresponding to each recognition parameter information according to the recognition result and the object area of ​​the extracted stored learning image, and obtain second recognition parameter information for second recognition processing based on the recognition threshold and the recognition number corresponding to each recognition parameter information;

[0073] The second recognition processing module is used to perform a second recognition process on the image to be recognized based on the second recognition parameter information to obtain a second recognition result, wherein the second recognition result includes a recognition area where the target object in the image to be recognized is located and a corresponding accuracy, and the second recognition result is saved.

[0074] The present invention also provides a computing device, the device comprising:

[0075] Memory and processor;

[0076] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the above-mentioned image analysis method based on big data technology is implemented.

[0077] According to another aspect of an embodiment of the present invention, a storage medium is further provided, wherein the storage medium stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute the above-mentioned image analysis method based on big data technology.

[0078] In summary, the image analysis method, system, device and storage medium based on big data technology of the present invention include presetting multiple recognition parameter information, using multiple recognition parameter information in sequence to perform first recognition processing on the learning image, and obtaining recognition results corresponding to different recognition parameter information; according to the recognition result and the target area of ​​the learning image stored in advance, obtaining the recognition threshold and recognition times corresponding to each recognition parameter information, and selecting the second recognition parameter information for the second recognition processing based on the recognition threshold and recognition times corresponding to each recognition parameter information; obtaining the image to be recognized, and performing the second recognition processing on the image to be recognized based on the second recognition parameter information to obtain the second recognition result. The present invention can improve the efficiency of the recognition model in identifying the target object from any image while ensuring the recognition accuracy.

[0079] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0080] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0081] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0082] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0083] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An image analysis method based on big data technology, characterized in that: The steps include: Pre-setting a plurality of recognition parameter information, the recognition parameter information including a recognition area size, a recognition area sliding size and a first threshold value, different recognition parameter information having different recognition area sliding sizes and first threshold values, sequentially using the plurality of recognition parameter information to perform a first recognition process on a learning image, the learning image being an image including a target object, pre-storing an object area of ​​the target object in the learning image, obtaining recognition results corresponding to different recognition parameter information, the recognition results including accuracy and a target area, the target area being an area where the target object is located; According to the recognition result and the object area of ​​the learning image stored in advance, obtaining a recognition threshold and a recognition number corresponding to each of the recognition parameter information, and obtaining second recognition parameter information for second recognition processing based on the recognition threshold and the recognition number corresponding to each of the recognition parameter information; Acquire an image to be identified, and perform a second identification process on the image to be identified based on the second identification parameter information to obtain a second identification result. Specifically, the second identification process step includes first acquiring the first second identification parameter information among all the second identification parameter information, and also acquiring the identification area, the identification area sliding size and the identification threshold corresponding to the second identification parameter information. According to the identification area size and the identification area sliding size, the identification area of ​​the image to be identified is acquired. The accuracy of each identification area is acquired using a recognition model. The accuracy is compared with the identification threshold of the second identification parameter information. If the accuracy is greater than or equal to the identification threshold, the corresponding identification area is used as the next area to be identified for the second identification parameter information. If the accuracy is less than the identification threshold, the identification area is excluded. The next second identification parameter information is acquired, and the previous step is repeated until the second identification process of all the second identification parameter information is completed. The second identification result includes the identification area where the target object in the image to be identified is located and the corresponding accuracy. The second identification result is saved.

2. The method according to claim 1, characterized in that Performing a first recognition process on the learning image includes the following steps: Obtain one piece of recognition parameter information from all the recognition parameter information, input the recognition parameter information and the learning image into a recognition model, set a recognition area size, the recognition model obtains all recognition areas of the learning image according to the recognition area size and the recognition area sliding size of the recognition parameter information, the recognition model identifies all the recognition areas and obtains a recognition result, the recognition result includes a target area and an accuracy of the target object in the recognition area, calculates an overlap rate between the object area and the target area according to the target area and an object area of ​​the target object in the learning image stored in advance, compares the overlap rate with a first threshold corresponding to the recognition parameter information, if the overlap rate is greater than or equal to the first threshold, determines that the recognition area contains the target object, if the overlap rate is less than the first threshold, determines that the recognition area does not contain the target object, obtains a total number of all recognition areas determined to contain the target object, and also obtains the accuracy of the corresponding recognition area, uses the total number as the number of recognitions of the recognition parameter information, obtains a minimum value in the accuracy, uses the minimum value as the recognition threshold of the recognition parameter information, and associates and saves the recognition threshold, the number of recognitions, and the recognition parameter information; Repeat the above steps to obtain the recognition times and corresponding recognition thresholds of all the recognition parameter information.

3. The method according to claim 1, characterized in that Acquiring second identification parameter information for second identification processing includes the following steps: Obtain the identification threshold and the number of identification times for each identification parameter information, add the number of identification times corresponding to all the identification parameter information to obtain a first total number, divide the first total number by a specific value to obtain an average number, the specific value is equal to the level number of the identification parameter information plus one, the level number of the identification parameter information is pre-defined when setting the identification parameter information, determine the best identification parameter information based on the average number, and use the best identification parameter information as the second identification parameter information.

4. The method according to claim 2, characterized in that Acquiring all recognition areas of the learning image according to the recognition area size and the recognition area sliding size of the recognition parameter information includes the following steps: Take the upper left corner of the learning image as the upper left corner of the recognition area to obtain a rectangular area with a width k and a height h of the recognition area size, take the rectangular area as the recognition area, and after the recognition model completes the recognition of the recognition area, obtain the sliding width and sliding height of the sliding size of the recognition area, move the rectangular area horizontally to the right by the distance of the sliding width, and obtain a new rectangular area each time it moves, and take the new rectangular area as the recognition area. After each horizontal movement to the right, determine whether the right boundary of the rectangular area coincides with the right boundary of the learning image. If not, continue to move horizontally to the right by the distance of the sliding width until the rectangular area The right boundary of the rectangular area coincides with the right boundary of the learning image, and then the rectangular area is moved downward in the vertical direction by the distance of the sliding height to obtain a new rectangular area, and the new rectangular area is identified as the recognition area, and then the rectangular area is moved leftward in the horizontal direction by the distance of the sliding width until the left boundary of the rectangular area coincides with the left boundary of the learning image, and then it is determined whether the lower boundary of the rectangular area coincides with the lower boundary of the learning image. If not, continue to move the rectangular area downward in the vertical direction by the distance of the sliding height, and repeat this step until the recognition processing is completed for all areas of the learning image. If they coincide, the processing is terminated.

5. The method according to claim 2, characterized in that: Calculating the overlap rate between the object area and the target area comprises the following steps: The overlap rate R is calculated using a first formula, which is: R=S3 / (S1+S2-S3), wherein S1 refers to the area of ​​the object region, S2 refers to the area of ​​the target region, and S3 refers to the area of ​​the overlapping region between the object region and the target region.

6. The image analysis method based on big data technology according to claim 1, characterized in that: After the recognition model is used to identify the recognition area and the accuracy of each recognition area is obtained, the following steps are performed: Obtaining the accuracy of each of the recognition areas, determining whether there is a recognition area whose accuracy is greater than or equal to the recognition threshold, and if so, marking the recognition area whose accuracy is greater than or equal to the recognition threshold. After determining all the recognition areas, selecting one of the marked recognition areas, setting a moving distance, moving the recognition area by the moving distance in each movable direction to obtain a new recognition area, identifying the new recognition area with the recognition model, obtaining the corresponding accuracy and object area, obtaining the accuracy of the new recognition area in each movable direction, and if the accuracy is greater than or equal to the recognition threshold, continuing to move the recognition area along the movable direction, performing recognition processing on the moved recognition area to obtain a corresponding recognition result, until the recognition area moves to the boundary of the image to be recognized, and terminating the processing; Determine whether there are other marked identification areas, and if so, repeat the previous step until all marked identification areas are identified.

7. An image analysis method system based on big data technology, used to implement the method according to any one of claims 1 to 6, characterized in that: Includes the following modules: A first recognition processing module is used to pre-set multiple recognition parameter information, the recognition parameter information includes a recognition area size, a recognition area sliding size and a first threshold, different recognition parameter information has different recognition area sliding sizes and corresponding first thresholds, and sequentially use the multiple recognition parameter information to perform first recognition processing on a learning image, the learning image refers to an image including a target object, pre-store an object area of ​​the target object in the learning image, and obtain recognition results corresponding to different recognition parameter information, the recognition results include accuracy and a target area, and the target area refers to an area where the target object is located; a recognition parameter selection module, configured to obtain a recognition threshold and a recognition number corresponding to each of the recognition parameter information according to the recognition result and the object area of ​​the extracted and stored learning image, and to obtain second recognition parameter information for second recognition processing based on the recognition threshold and the recognition number corresponding to each of the recognition parameter information; A second recognition processing module is used to perform a second recognition processing on the image to be recognized based on the second recognition parameter information to obtain a second recognition result. The specific second recognition processing steps include first obtaining the first second recognition parameter information among all the second recognition parameter information, and also obtaining the recognition area, recognition area sliding size and recognition threshold corresponding to the second recognition parameter information; obtaining the recognition area of ​​the image to be recognized according to the recognition area size and the recognition area sliding size; using a recognition model to obtain the accuracy of each of the recognition areas; comparing the accuracy with the recognition threshold of the second recognition parameter information; if the accuracy is greater than or equal to the recognition threshold, using the corresponding recognition area as the next area to be recognized for the second recognition parameter information; if the accuracy is less than the recognition threshold, excluding the recognition area; obtaining the next second recognition parameter information, repeating the previous step until the second recognition processing of all the second recognition parameter information is completed; the second recognition result includes the recognition area where the target object in the image to be recognized is located and the corresponding accuracy; and the second recognition result is saved.

8. An electronic device, characterized in that: include: A processor, a memory and a communication bus, wherein the processor and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is used to execute the program stored in the memory to implement the method described in any one of claims 1-6.

9. A storage medium, characterized in that: The storage medium stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 6.

Images