CN112381223A

CN112381223A - Neural network training and image processing method and device

Info

Publication number: CN112381223A
Application number: CN202011269792.4A
Authority: CN
Inventors: 李嘉辉; 陈文�; 黄晓迪; 胡志强; 张少霆
Original assignee: Shanghai Sensetime Intelligent Technology Co Ltd
Current assignee: Shanghai Sensetime Intelligent Technology Co Ltd
Priority date: 2020-11-13
Filing date: 2020-11-13
Publication date: 2021-02-19

Abstract

The present disclosure relates to a neural network training and image processing method and apparatus, which can be used for labeling medical image images, such as heart, liver, lung and pathological images. The method comprises the following steps: training the neural network according to the first sample image to obtain the neural network in a first training state; and training the neural network in the first training state according to the second sample image and the first sample image to obtain the neural network in the second training state. According to the neural network training method disclosed by the embodiment of the disclosure, the neural network can be gradually trained through the sample images with different labels, so that the capability of the neural network for processing the sample images with different labels is enhanced, and the precision and the generalization capability of the neural network are improved.

Description

Neural network training and image processing method and device

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a neural network training and image processing method and apparatus.

Background

In the related art, labeling accuracy differs for samples, and for example, according to the labeling accuracy, the samples can be classified into samples without labeling, samples with class labeling, samples with detection frame labeling, and samples with segmentation contour labeling. If the labeled information amount is different, the labeling cost is also different, for example, the information amount of the segmentation contour label or the detection frame label is greater than the category label, and the labeling cost is also greater than the category label. Thus, the number of samples with segmentation contour labeling is less than the number of samples with class labeling. In the neural network training process, the number of samples with large labeled information amount is small, so that the precision and the generalization capability of the trained neural network are insufficient. And training by using a sample with small labeling information amount, so that the performance of the trained neural network is insufficient.

Disclosure of Invention

The disclosure provides a neural network training and image processing method and device.

According to an aspect of the present disclosure, there is provided a neural network training method, including: training a neural network according to a first sample image to obtain the neural network in a first training state, wherein the first sample image has region labeling information; and training the neural network in the first training state according to a second sample image and the first sample image to obtain the neural network in a second training state, wherein the second sample image has class marking information.

According to the neural network training method disclosed by the embodiment of the disclosure, the neural network can be gradually trained through the sample images with different labels, so that the capability of the neural network for processing the sample images with different labels is enhanced, and the precision and the generalization capability of the neural network are improved.

In a possible implementation manner, training the neural network in the first training state according to the second sample image and the first sample image to obtain the neural network in the second training state includes: detecting and processing a second sample image by using a neural network of an ith training period to obtain a first prediction region in the second sample image, wherein i is a positive integer, and the ith training period is any one of a plurality of training periods for training the neural network in the first training state into the neural network in the second training state; training the neural network of the ith training period by using a second sample image with class labeling information and the first prediction region and a first sample image with region labeling information to obtain a neural network of an (i + 1) th training period; and when the neural network meets a first training condition, obtaining the neural network in the second training state.

In this way, the neural network can be trained by the second sample image with the class marking information, the number of the first sample images is supplemented, and the detection precision of the neural network is improved.

In a possible implementation manner, performing detection processing on a second sample image with class labeling information by using a neural network in an ith training cycle to obtain a first prediction region in the second sample image, includes: detecting and processing a second sample image with class marking information by using the neural network of the ith training period to obtain a first sample area in the second sample image; determining a second sample region with a confidence level greater than or equal to a first confidence level threshold in the first sample region; and determining the first prediction region in the second sample region according to the class marking information.

In one possible implementation, the method further includes: and training the neural network in the second training state according to the third sample image, the second sample image and the first sample image which are not marked to obtain the trained neural network.

By the method, the neural network can be trained by a large number of unmarked third sample images and the second sample images with the class marking information, so that the marking cost of the sample images is reduced.

In a possible implementation manner, training the neural network in the second training state according to the unlabeled third sample image, the second sample image, and the first sample image, to obtain a trained neural network, includes: detecting and processing an unlabelled third sample image by using a neural network of a jth training period to obtain a second prediction region in the third sample image, wherein j is a positive integer, and the jth training period is any one of a plurality of training periods for training the neural network in the second training state into a trained neural network; training the neural network of the jth training period by using a first sample image with region labeling information, a second sample image with category labeling information and the first prediction region and a third sample image with a second prediction region to obtain a neural network of a (j + 1) th training period; and when the neural network meets a second training condition, obtaining the trained neural network.

In this way, the neural network can be trained through the third sample image which is not labeled, the detection precision of the neural network is further improved, and the third sample image which is not labeled with the category labeling information or the region labeling information is used for training the neural network, so that the robustness and the generalization capability of the neural network can be improved.

In a possible implementation manner, performing detection processing on an unlabeled third sample image by using a neural network in a jth training cycle to obtain a second prediction region in the third sample image, includes: detecting the third sample image by using the neural network of the jth training cycle to obtain a third sample region in the third sample image; in the third sample region, a second prediction region is determined with a confidence greater than or equal to a second confidence threshold.

In one possible implementation, in the third sample region, determining a second prediction region with a confidence greater than or equal to a second confidence threshold includes: determining a confidence level for the third sample region based on the size of the third sample region; determining the second prediction region according to the confidence of the third sample region.

In one possible implementation, the number of the first sample images is smaller than the number of the second sample images, and the number of the second sample images is smaller than the number of the third sample images.

In one possible implementation, the first sample image, the second sample image, and the third sample image are medical images.

According to an aspect of the present disclosure, there is provided an image processing method including: and processing the image to be processed according to the neural network trained by the neural network training method to obtain the position information and the category information of the target area in the image to be processed.

According to an aspect of the present disclosure, there is provided a neural network training apparatus including: the first training module is used for training the neural network according to a first sample image to obtain the neural network in a first training state, wherein the first sample image has region marking information; and the second training module is used for training the neural network in the first training state according to a second sample image and the first sample image to obtain the neural network in a second training state, wherein the second sample image has class marking information.

In one possible implementation, the second training module is further configured to: detecting and processing a second sample image by using a neural network of an ith training period to obtain a first prediction region in the second sample image, wherein i is a positive integer, and the ith training period is any one of a plurality of training periods for training the neural network in the first training state into the neural network in the second training state; training the neural network of the ith training period by using a second sample image with class labeling information and the first prediction region and a first sample image with region labeling information to obtain a neural network of an (i + 1) th training period; and when the neural network meets a first training condition, obtaining the neural network in the second training state.

In one possible implementation, the second training module is further configured to: detecting and processing a second sample image with class marking information by using the neural network of the ith training period to obtain a first sample area in the second sample image; determining a second sample region with a confidence level greater than or equal to a first confidence level threshold in the first sample region; and determining the first prediction region in the second sample region according to the class marking information.

In one possible implementation, the apparatus further includes a third training module configured to: and training the neural network in the second training state according to the third sample image, the second sample image and the first sample image which are not marked to obtain the trained neural network.

In one possible implementation, the third training module is further configured to: detecting and processing an unlabelled third sample image by using a neural network of a jth training period to obtain a second prediction region in the third sample image, wherein j is a positive integer, and the jth training period is any one of a plurality of training periods for training the neural network in the second training state into a trained neural network; training the neural network of the jth training period by using a first sample image with region labeling information, a second sample image with category labeling information and the first prediction region and a third sample image with a second prediction region to obtain a neural network of a (j + 1) th training period; and when the neural network meets a second training condition, obtaining the trained neural network.

In one possible implementation, the third training module is further configured to: detecting the third sample image by using the neural network of the jth training cycle to obtain a third sample region in the third sample image; in the third sample region, a second prediction region is determined with a confidence greater than or equal to a second confidence threshold.

In one possible implementation, the third training module is further configured to: determining a confidence level for the third sample region based on the size of the third sample region; determining the second prediction region according to the confidence of the third sample region.

In a possible implementation manner, the present disclosure further provides an image processing apparatus, including a processing module, configured to process an image to be processed according to a neural network trained by the neural network training method, and obtain location information and category information of a target area in the image to be processed.

According to an aspect of the present disclosure, there is provided an electronic device including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: the above method is performed.

According to an aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the above-described method.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

FIG. 1 shows a flow diagram of an image processing method according to an embodiment of the present disclosure;

FIG. 2 shows a schematic diagram of an application of a neural network training method according to an embodiment of the present disclosure;

FIG. 3 shows a block diagram of a neural network training device, in accordance with an embodiment of the present disclosure;

FIG. 4 shows a block diagram of an electronic device in accordance with an embodiment of the disclosure;

fig. 5 shows a block diagram of an electronic device according to an embodiment of the disclosure.

Detailed Description

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, C, and may mean including any one or more elements selected from the group consisting of A, B and C.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.

Fig. 1 shows a flowchart of an image processing method according to an embodiment of the present disclosure, as shown in fig. 1, the method comprising:

in step S11, training a neural network according to a first sample image, to obtain a neural network in a first training state, where the first sample image has region labeling information;

in step S12, the neural network in the first training state is trained according to a second sample image and the first sample image, so as to obtain a neural network in a second training state, where the second sample image has class label information.

In one possible implementation, the neural network training method may be performed by a terminal device or other processing device, where the terminal device may be a User Equipment (UE), a mobile device, a User terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or the like. The other processing devices may be servers or cloud servers, etc. In some possible implementations, the neural network training method may be implemented by a processor invoking computer readable instructions stored in a memory.

In a possible implementation manner, the first sample image may have region labeling information on the target region, for example, the position of the target region may be labeled by a rectangular detection frame, or the contour of the target region may be labeled by a segmentation contour line. The second sample image may have category labeling information for a category, for example, the category of the target object in the second sample image may be labeled, but the position of the target area where the target object is located is not labeled. The third sample image may not have any annotation information.

In an example, the first, second, and third sample images may be medical images in which the location of the lesion region may be marked by a detection frame or a dividing line. In the second sample image, a category of the lesion may be marked. The third sample image may not have any annotation information.

In an example, since the labeling difficulty of the detection frame or the segmentation line of the first sample image is high, the cost is high, and the number of the first sample images is generally small, for example, the first sample image may be a medical image, and a professional such as a doctor is generally required to identify the category and the region of the lesion and label the lesion.

In an example, the second sample image has category label information. The lesion classification can be identified and labeled, typically by a professional such as a physician. The labeling difficulty and cost are lower than those of the first sample image, and the number of the second sample images can be more than that of the first sample images. For example, a patient takes a plurality of medical images of the same cause, that is, the plurality of medical images of the same cause may have category label information (that is, the categories of the plurality of medical images of the same cause may be labeled as lesion images), but a doctor performs region labeling on only one of the partial images (for example, a target region is selected by a frame or a contour of the target region is marked by a dividing line), that is, only one of the partial images has position label information (the first sample image may have both category label information and region label information), and thus the number of the second sample images is greater than that of the first sample image.

In an example, the third sample image may not have labeling information, e.g., the third sample image may be a medical image labeled for a doctor, or a medical image derived from a network. The number of third sample images may be greater than the second sample images due to no annotation cost. In summary, the number of the first sample images is smaller than the number of the second sample images, and the number of the second sample images is smaller than the number of the third sample images.

In one possible implementation, in step S11, the neural network may be trained using the first sample image, obtaining a neural network in a first training state. The first sample image may have region labeling information, and after training, the neural network in the first training state may have the capability of distinguishing the type of the input image (e.g., the type of the lesion) and the region where the detection target object is located (e.g., the region where the lesion is located).

In one possible implementation, in step S12, to further improve the accuracy of the neural network, the neural network may be further trained, and in the absence of the first sample image with the region labeling information, the neural network may be trained using the second sample image with the category labeling information.

In one possible implementation, step S12 may include: detecting and processing a second sample image by using a neural network of an ith training period to obtain a first prediction region in the second sample image, wherein i is a positive integer, and the ith training period is any one of a plurality of training periods for training the neural network in the first training state into the neural network in the second training state; training the neural network of the ith training period by using a second sample image with class labeling information and the first prediction region and a first sample image with region labeling information to obtain a neural network of an (i + 1) th training period; and when the neural network meets a first training condition, obtaining the neural network in the second training state.

In one possible implementation manner, the neural network in the first training state may be trained for a plurality of training cycles according to the second sample image, so as to obtain the neural network in the second training state. For example, a neural network in a first training state (a first training period) may be used to detect a second sample image having only category label information, and a first prediction region (a prediction region with higher accuracy after being filtered) in the second sample image may be obtained, that is, the second sample image having the first prediction region may be obtained. Further, the neural network of the first training period may be obtained by training the neural network of the first training state using the second sample image having the first prediction region and the first sample image having the region labeling information, or by training a new neural network using the second sample image having the first prediction region and the first sample image having the region labeling information (since the total number of the second sample image having the first prediction region and the first sample image is larger than that of the first sample image, a new neural network is trained using the two sample images, and after training, the accuracy of the neural network is higher than that of the neural network of the first training state).

In an example, taking any one of the training periods (i-th training period) as an example, the neural network of the i-th training period may be used to detect the second sample image with only the category label information, so that the first prediction region (the prediction region with higher accuracy after being filtered) in the second sample image may be obtained, that is, the second sample image with the first prediction region may be obtained. Further, the neural network of the i-th training cycle may be trained using the second sample image with the first prediction region and the first sample image with the region-labeling information obtained in the 1 st to i-th training cycles, or the neural network of the i + 1-th training cycle may be obtained by training a new neural network using the second sample image with the first prediction region and the first sample image with the region-labeling information obtained in the 1 st to i-th training cycles.

In one possible implementation, an error may occur when the prediction region in the second sample image is determined by the neural network of an arbitrary training period. Therefore, the prediction region determined by the neural network may be screened to determine a prediction region with higher accuracy, i.e., the first prediction region.

In an example, taking an ith training cycle as an example, the neural network of the ith training cycle may perform detection processing on the second sample image with the category label information to obtain a first sample region in the second sample image, and there may be an error in obtaining the first sample region through neural network prediction, for example, a non-target region may be predicted as the first sample region, or a target region may be missed to be detected.

In an example, in the obtained first sample regions, each first sample region may have a corresponding confidence. For example, the neural network may determine the first sample region by determining a probability that each pixel belongs to the first sample region (e.g., if a probability that a certain pixel belongs to the first sample region is greater than 50%, the certain pixel may be determined as a pixel in the first sample region, and a plurality of pixels belonging to the first sample region may form the first sample region). The confidence of the first sample region can be determined by the probability of each pixel point in the first sample region corresponding, and a second sample region with a confidence greater than or equal to a first confidence threshold can be determined. For example, in the first sample region, the confidence of a part of the region is high, that is, the accuracy of the neural network judging the part of the region as the target region is high. And the confidence of the other part of the region is lower, namely, the accuracy of judging the part of the region as the target region by the neural network is lower, the region with lower confidence can be deleted, and the region with the confidence greater than or equal to the first confidence threshold value, namely, the second sample region, is reserved.

In an example, there may be a certain confidence range for the size of the first sample region in the example, i.e. the confidence is higher for the size of the region of the first sample within a certain interval and lower for the first sample region if the size exceeds that interval. For example, the first sample region is a frame-shaped region, and the length of the first sample region is within a preset length interval, so that the confidence of the first sample region is higher. For example, if the width of the first sample region is within the preset width interval, the confidence of the first sample region is higher. For another example, if the area of the first sample region is within the preset area interval, the confidence of the first sample region is higher. The regions of lower confidence may be deleted, leaving regions with confidence greater than or equal to the first confidence threshold, i.e., the second sample region.

In an example, the second sample region can be further filtered according to the category label information to obtain the first prediction region. For example, the second sample image is a medical image, and the category label information of the medical image may include a lesion image (e.g., including a lesion region in the image) and a non-lesion image (e.g., not including a lesion region in the image). The second sample region may be screened using the class label information of the second sample image, for example, if the class label information of the second sample image is a non-lesion image, the second sample regions (lesion regions) predicted by the neural network are all wrong prediction results, that is, if the second sample image is a lung lesion image in which a lesion region is not included, the lesion regions predicted by the neural network are wrong, and thus, these lesion regions may be deleted. For another example, if the category label information of the second sample image is a lesion image, the second sample regions (lesion regions) predicted by the neural network are all correct prediction results, in which case, the confidence of the second sample regions may be doubled to improve the confidence of the correct prediction results, and the second sample regions may be determined as the first prediction regions (i.e., prediction regions with higher accuracy). The neural network of the i-th training period may be trained through the second sample image having the first prediction region and the first sample image to obtain the neural network of the i + 1-th period.

In one possible implementation, the training process described above may be iteratively performed a plurality of times until the first training condition is satisfied. The first training condition may include that the accuracy of the neural network meets the accuracy requirement, for example, when the neural network is used to detect the second sample image, the prediction accuracy is greater than or equal to a preset accuracy threshold (for example, the proportion of prediction regions with confidence higher than the first confidence threshold in the obtained prediction regions is greater than or equal to the preset accuracy threshold), and the neural network in the second training state may be obtained. The first training condition may further include a training number, for example, the training process may be performed iteratively for a preset number of times, or after the second sample image is completely trained, a neural network of the second training state may be obtained, and the disclosure does not limit the first training condition.

In a possible implementation manner, the neural network can be further trained through the third sample image without the label, so that the performance of the neural network is improved. The method further comprises the following steps: and training the neural network in the second training state according to the third sample image, the second sample image and the first sample image which are not marked to obtain the trained neural network.

In one possible implementation, the foregoing steps may include: detecting and processing an unlabelled third sample image by using a neural network of a jth training period to obtain a second prediction region in the third sample image, wherein j is a positive integer, and the jth training period is any one of a plurality of training periods for training the neural network in the second training state into a trained neural network; training the neural network of the jth training period by using a first sample image with region labeling information, a second sample image with category labeling information and the first prediction region and a third sample image with a second prediction region to obtain a neural network of a (j + 1) th training period; and when the neural network meets a second training condition, obtaining the trained neural network.

In a possible implementation manner, the neural network in the second training state may be trained for a plurality of training cycles according to the third sample image, so as to obtain a trained neural network. For example, the unlabeled third sample image may be detected using the neural network in the second training state (the first training period), and the second prediction region (the prediction region with higher accuracy after being filtered) in the third sample image may be obtained, that is, the third sample image with the second prediction region may be obtained. Further, a neural network of the second training period may be trained using a third sample image having the second prediction region, a second sample image having the first prediction region, and a first sample image having region labeling information, or a new neural network may be trained using the sample images (since the total number of the sample images is greater than the total number of the first sample image and the second sample image having the first prediction region, a new neural network trained using the three sample images, the neural network having a higher accuracy than the neural network of the second training state after training), and the neural network of the second training period may be obtained.

In an example, taking any one of the training periods (jth training period) as an example, the neural network of the jth training period may be used to detect the unlabeled third sample image, so as to obtain a second prediction region (a prediction region with higher accuracy after being screened) in the third sample image, that is, obtain the third sample image with the second prediction region. Further, the neural network of the jth training cycle may be trained using the third sample image having the second prediction region, the second sample image having the first prediction region, and the first sample image having the region labeling information obtained in the 1 st to jth training cycles, or may be trained using the above sample images to train a new neural network, and the neural network of the j +1 th training cycle may be obtained.

In one possible implementation, an error may occur when the prediction region in the third sample image is determined by the neural network of an arbitrary training period. Therefore, the prediction region determined by the neural network may be screened to determine a prediction region with higher accuracy, i.e., the second prediction region.

In an example, taking the jth training cycle as an example, the neural network of the jth training cycle may perform detection processing on the unlabeled third sample image, obtain a third sample region in the third sample image, and predict, by the neural network, that the third sample region may have an error, for example, a non-target region may be predicted as the third sample region, or a target region may be missed.

In an example, among the obtained third sample regions, each third sample region may have a corresponding confidence. For example, the neural network may determine the third sample region by determining a probability that each pixel belongs to the third sample region (e.g., if a probability that a certain pixel belongs to the third sample region is greater than 50%, the certain pixel may be determined as a pixel in the third sample region, and a plurality of pixels belonging to the third sample region may constitute the third sample region). The confidence of the third sample region can be determined by the probability of each pixel point in the third sample region corresponding, and a second prediction region with a confidence greater than or equal to a second confidence threshold can be determined. For example, in the third sample region, the confidence of a part of the region is high, that is, the accuracy of the neural network judging the part of the region as the target region is high. And the confidence of the other part of the region is lower, namely, the accuracy of judging the part of the region as the target region by the neural network is lower, the region with lower confidence can be deleted, and the region with the confidence greater than or equal to the second confidence threshold value is reserved, namely, the second prediction region.

In an example, in the third sample region, determining a second prediction region with a confidence greater than or equal to a second confidence threshold comprises: determining confidence levels of a plurality of pixel points in a third sample region according to the size of the third sample region; and determining the second prediction region according to the confidence degrees of a plurality of pixel points in the third sample region.

In an example, there may be a range of confidence in the size of the region of the third sample, i.e., the confidence is higher if the size of the region of the third sample is within a certain interval, and the confidence is lower if the size exceeds that interval. For example, if the third sample region is a frame-shaped region and the length of the third sample region is within the preset length interval, the confidence of the third sample region is higher. For example, if the width of the third sample region is within the preset width interval, the confidence of the third sample region is higher. For another example, if the area of the third sample region is within the preset area interval, the confidence of the third sample region is higher. Regions with lower confidence may be deleted, leaving regions with confidence greater than or equal to a second confidence threshold, i.e., second prediction regions.

In one possible implementation, the training process described above may be iteratively performed a plurality of times until a second training condition is satisfied. The second training condition may include that the accuracy of the neural network meets the accuracy requirement, for example, when the neural network is used to detect the third sample image, the prediction accuracy is greater than or equal to a preset accuracy threshold (for example, the proportion of prediction regions with confidence degrees higher than the second confidence degree threshold in the obtained prediction regions is greater than or equal to the preset accuracy threshold), and the trained neural network may be obtained. The second training condition may further include a training number, for example, the training process may be performed iteratively for a preset number of times, or after the third sample image is completely trained, a trained neural network may be obtained, and the disclosure does not limit the second training condition.

In a possible implementation manner, the trained neural network has high detection accuracy and generalization capability, and can identify the type of the image and the position of the target region, for example, a medical image can be input into the neural network, whether the medical image is a focus image (the type of the medical image is identified) can be obtained, and if the medical image is the focus image, the focus region can be detected.

In one possible implementation, the present disclosure further provides an image processing method, including: and processing the image to be processed by the neural network trained according to the neural network training method to obtain the position information and the category information of the target area in the image to be processed.

In an example, the image to be processed may be a medical image. The image to be processed may be input into the trained neural network for processing, so as to obtain category information of the image to be processed, for example, whether the image to be processed is a lesion image (the image to be processed includes a lesion region) or a non-lesion image (the image to be processed does not include a lesion region). If the category information of the image to be processed is a lesion image, the position information of the target region (lesion region) may be further detected, for example, the target region may be selected by a rectangular detection frame, or the contour of the target region may be segmented by segmenting a contour line.

According to the neural network training method disclosed by the embodiment of the disclosure, the neural network can be trained through the second sample image with the class marking information, the number of the first sample images is supplemented, and the detection precision of the neural network is improved. And the third sample image without the category labeling information and the region labeling information is used for training the neural network, so that the robustness and the generalization capability of the neural network can be improved. The neural network is gradually trained through the sample images with different labels, so that the labeling cost can be reduced, and the performance of the neural network can be enhanced.

Fig. 2 is a schematic diagram illustrating an application of a neural network training method according to an embodiment of the present disclosure, and as shown in fig. 2, a first sample image, a second sample image and a third sample image may be medical images, the first sample image may have region labeling information (e.g., a segmentation contour or a rectangular frame) on a lesion region, the second sample image may have category labeling information (e.g., whether the sample image is a lesion image), and the third sample image may have no label. And the number of the first sample images is less than that of the second sample images, and the number of the second sample images is less than that of the third sample images.

In one possible implementation, the neural network may be trained using a smaller number of first sample images, resulting in a neural network in a first training state. The accuracy of the neural network in the first training state is not high due to the small number of first sample images. Training may continue through the second sample image.

In a possible implementation manner, the second sample image has no region labeling information and cannot be directly used for training. The second sample image can be processed through the neural network in the first training state to obtain the first sample region, and the first sample region may have errors because the neural network in the first training state is not high in precision. The first sample region can be screened, for example, a second sample region with a confidence level above a first confidence threshold can be screened from the plurality of first sample regions. Further, the second sample regions may be continuously screened according to the category label information, for example, if the category label information of the second sample image is a non-lesion image, the second sample regions (lesion regions) predicted by the neural network are all wrong prediction results, and these lesion regions may be deleted, and for example, if the category label information of the second sample image is a lesion image, the second sample regions (lesion regions) predicted by the neural network are all correct prediction results, in which case, the confidence of the second sample regions may be doubled to obtain the first prediction regions. And training of the neural network may continue with the second sample image having the first prediction region and the first sample image. The training process described above may be iteratively performed until the first training condition is satisfied, resulting in a neural network in a second training state.

In one possible implementation, the neural network may continue to be trained through the third unlabeled sample image to continue to improve the accuracy and robustness of the neural network. The third sample image may be processed using the neural network in the second training state to obtain a third sample region, wherein the third sample region may have errors. The third sample region may be screened, for example, a second prediction region with a confidence greater than or equal to a second confidence threshold may be screened from the third sample region and the confidence of the second prediction region may be doubled, and a third sample region with a confidence less than the second confidence threshold may be deleted. Further, training of the neural network may continue with a third sample image having the second prediction region, a second sample image having the first prediction region, and the first sample image. The training process may be iteratively performed until a second training condition is satisfied, resulting in a trained neural network.

In one possible implementation manner, the category information of the image to be processed and the position information of the lesion area may be obtained through the trained neural network. For example, the image to be processed may be input to a neural network. The category information of the image to be processed is obtained, for example, whether the image to be processed is a lesion image (including a lesion region in the image to be processed) or a non-lesion image (not including a lesion region in the image to be processed). If the category information of the image to be processed is a lesion image, the position information of the target region (lesion region) can be further detected.

In a possible implementation manner, the neural network training method can be used in the field of image processing, and if the sample labeling cost is high, for example, the labeling cost for medical images is high, the neural network can be gradually trained by samples with different labels, so that the neural network has high precision and robustness and can be used in image detection and recognition and other processing. The application field of the neural network training method is not limited by the disclosure.

Fig. 3 shows a block diagram of a neural network training device according to an embodiment of the present disclosure, as shown in fig. 3, the device including: the first training module 11 is configured to train a neural network according to a first sample image to obtain the neural network in a first training state, where the first sample image has region labeling information; the second training module 12 is configured to train the neural network in the first training state according to a second sample image and the first sample image, to obtain the neural network in a second training state, where the second sample image has class label information.

It is understood that the above-mentioned method embodiments of the present disclosure can be combined with each other to form a combined embodiment without departing from the logic of the principle, which is limited by the space, and the detailed description of the present disclosure is omitted.

In addition, the present disclosure also provides a neural network training device, an electronic device, a computer-readable storage medium, and a program, which can be used to implement any one of the neural network training methods provided by the present disclosure, and the corresponding technical solutions and descriptions and corresponding descriptions in the methods section are not repeated.

It will be understood by those skilled in the art that in the method of the present invention, the order of writing the steps does not imply a strict order of execution and any limitations on the implementation, and the specific order of execution of the steps should be determined by their function and possible inherent logic.

In some embodiments, functions of or modules included in the apparatus provided in the embodiments of the present disclosure may be used to execute the method described in the above method embodiments, and for specific implementation, reference may be made to the description of the above method embodiments, and for brevity, details are not described here again

Embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the above-mentioned method. The computer readable storage medium may be a non-volatile computer readable storage medium.

An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured as the above method.

The electronic device may be provided as a terminal, server, or other form of device.

Fig. 4 is a block diagram illustrating an electronic device 800 in accordance with an example embodiment. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like terminal.

Referring to fig. 4, electronic device 800 may include one or more of the following components: processing component 802, memory 804, power component 806, multimedia component 808, audio component 810, input/output (I/O) interface 812, sensor component 814, and communication component 816.

The processing component 802 generally controls overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing components 802 may include one or more processors 820 to execute instructions to perform all or a portion of the steps of the methods described above. Further, the processing component 802 can include one or more modules that facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.

The memory 804 is configured to store various types of data to support operations at the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 804 may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

The power supply component 806 provides power to the various components of the electronic device 800. The power components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the electronic device 800 is in an operation mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 810 is configured to output and/or input audio signals. For example, the audio component 810 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 800 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 804 or transmitted via the communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing component 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly 814 includes one or more sensors for providing various aspects of state assessment for the electronic device 800. For example, the sensor assembly 814 may detect an open/closed state of the electronic device 800, the relative positioning of components, such as a display and keypad of the electronic device 800, the sensor assembly 814 may also detect a change in the position of the electronic device 800 or a component of the electronic device 800, the presence or absence of user contact with the electronic device 800, orientation or acceleration/deceleration of the electronic device 800, and a change in the temperature of the electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 800 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 804, is also provided that includes computer program instructions executable by the processor 820 of the electronic device 800 to perform the above-described methods.

Fig. 5 is a block diagram illustrating an electronic device 1900 according to an example embodiment. For example, the electronic device 1900 may be provided as a server. Referring to fig. 5, electronic device 1900 includes a processing component 1922 further including one or more processors and memory resources, represented by memory 1932, for storing instructions, e.g., applications, executable by processing component 1922. The application programs stored in memory 1932 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1922 is configured to execute instructions to perform the above-described method.

The electronic device 1900 may also include a power component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input/output (I/O) interface 1958. The electronic device 1900 may operate based on an operating system stored in memory 1932, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.

In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1932, is also provided that includes computer program instructions executable by the processing component 1922 of the electronic device 1900 to perform the above-described methods.

The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A neural network training method, comprising:

training a neural network according to a first sample image to obtain the neural network in a first training state, wherein the first sample image has region labeling information;

and training the neural network in the first training state according to a second sample image and the first sample image to obtain the neural network in a second training state, wherein the second sample image has class marking information.

2. The method of claim 1, wherein training the neural network in the first training state according to a second sample image and the first sample image to obtain the neural network in a second training state comprises:

detecting and processing a second sample image by using a neural network of an ith training period to obtain a first prediction region in the second sample image, wherein i is a positive integer, and the ith training period is any one of a plurality of training periods for training the neural network in the first training state into the neural network in the second training state;

training the neural network of the ith training period by using a second sample image with class labeling information and the first prediction region and a first sample image with region labeling information to obtain a neural network of an (i + 1) th training period;

and when the neural network meets a first training condition, obtaining the neural network in the second training state.

3. The method according to claim 2, wherein the detecting process is performed on the second sample image with the class labeling information by using a neural network of an ith training cycle to obtain the first prediction region in the second sample image, and the method comprises:

detecting and processing a second sample image with class marking information by using the neural network of the ith training period to obtain a first sample area in the second sample image;

determining a second sample region with a confidence level greater than or equal to a first confidence level threshold in the first sample region;

and determining the first prediction region in the second sample region according to the class marking information.

4. The method of claim 1, further comprising:

and training the neural network in the second training state according to the third sample image, the second sample image and the first sample image which are not marked to obtain the trained neural network.

5. The method of claim 4, wherein training the neural network in the second training state according to the unlabeled third sample image, the second sample image, and the first sample image to obtain a trained neural network comprises:

detecting and processing an unlabelled third sample image by using a neural network of a jth training period to obtain a second prediction region in the third sample image, wherein j is a positive integer, and the jth training period is any one of a plurality of training periods for training the neural network in the second training state into a trained neural network;

training the neural network of the jth training period by using a first sample image with region labeling information, a second sample image with category labeling information and the first prediction region and a third sample image with a second prediction region to obtain a neural network of a (j + 1) th training period;

and when the neural network meets a second training condition, obtaining the trained neural network.

6. The method of claim 5, wherein the detecting the unlabeled third sample image using the neural network of the jth training cycle to obtain the second prediction region in the third sample image comprises:

detecting the third sample image by using the neural network of the jth training cycle to obtain a third sample region in the third sample image;

in the third sample region, a second prediction region is determined with a confidence greater than or equal to a second confidence threshold.

7. The method of claim 6, wherein determining, in the third sample region, a second prediction region having a confidence greater than or equal to a second confidence threshold comprises:

determining a confidence level for the third sample region based on the size of the third sample region;

determining the second prediction region according to the confidence of the third sample region.

8. The method according to any one of claims 1 to 7, wherein the number of the first sample images is smaller than the number of second sample images, which is smaller than the number of third sample images.

9. The method according to any one of claims 1-8, wherein the first, second and third sample images are medical images.

10. An image processing method, comprising:

the neural network trained by the neural network training method according to any one of claims 1 to 9, processing the image to be processed to obtain the position information and the category information of the target region in the image to be processed.

11. A neural network training device, comprising:

the first training module is used for training the neural network according to a first sample image to obtain the neural network in a first training state, wherein the first sample image has region marking information;

and the second training module is used for training the neural network in the first training state according to a second sample image and the first sample image to obtain the neural network in a second training state, wherein the second sample image has class marking information.

12. An image processing apparatus characterized by comprising:

a processing module, configured to process the image to be processed according to the neural network trained by the neural network training method according to any one of claims 1 to 9, so as to obtain location information and category information of the target area in the image to be processed.

13. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to: performing the method of any one of claims 1 to 10.

14. A computer readable storage medium having computer program instructions stored thereon, which when executed by a processor implement the method of any one of claims 1 to 10.