CN111898680A - Biological identification method based on material inspection multi-view morphological image and deep learning - Google Patents
Biological identification method based on material inspection multi-view morphological image and deep learning Download PDFInfo
- Publication number
- CN111898680A CN111898680A CN202010759848.8A CN202010759848A CN111898680A CN 111898680 A CN111898680 A CN 111898680A CN 202010759848 A CN202010759848 A CN 202010759848A CN 111898680 A CN111898680 A CN 111898680A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- convolutional neural
- biological
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5854—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using shape and object relationship
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
Abstract
The invention discloses a biological identification method based on material inspection multi-view morphological images and deep learning, which comprises the following steps: establishing a model, detecting material sampling, extracting characteristics until extracting image characteristics corresponding to external form images acquired under different visual angles, combining the characteristics, comparing and identifying: compared with the prior art, the identification method provided by the invention has the beneficial effects of labor saving, high identification efficiency and high accuracy.
Description
Technical Field
The invention belongs to the technical field of image recognition, and particularly relates to a species identification method of a multi-dimensional image based on organisms.
Background
In the fields of import and export inspection and quarantine, natural environment scientific investigation, judicial identification and the like, identification is often required to be carried out on the material to be detected, and the specific species information of organisms contained in the material to be detected is identified.
In the prior art, in order to realize biological species identification, methods such as morphology, serology, cytology, biochemistry and the like are usually adopted, the methods have high requirements on the quality of a detected material, the whole preparation process of the detected product is quite complex, the requirements on preparation equipment, preparation environment and preparation personnel are particularly high, most of the whole identification process is manual operation, the whole operation flow is quite complex, the repeatability is extremely high, the manual dependence degree is quite high, and the efficiency of the whole identification process is quite low.
In the occasions of import and export inspection and quarantine, natural environment scientific investigation, judicial identification and the like, the situations that the inspection materials are biological specimens with complete appearances are very common, when the inspection materials are identified, an identifier needs to rely on the prior identification experience, look up professional data such as a biological search table or a species graph identification and the like, and can finally obtain a species identification result after comparison and search, the process very examines the knowledge reserve of the identifier, and the whole process has strong experience.
In recent years, with the development of computer technology, deep learning with its high autonomy, wide adaptability and strong computing power is applied in more and more fields, in particular to the field of biological identification, how to judge the specific biological species of a material to be tested with complete external morphology according to the external morphology based on the deep learning technology, and no relevant data record has been found before. Therefore, how to apply the deep learning technique to the bioassay process by using the external morphology of the sample to obtain an ideal assay result is a technical problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In order to solve the above problems, an object of the present invention is to provide a method for identifying species of a material to be inspected having a complete external shape by using an external shape image of the material based on a deep learning technique.
Another objective of the present invention is to provide a biological identification method based on the exterior shape image of the material under test, which obtains the exterior shape image of the material under test from multiple viewing angles, extracts the image features of each image, and combines the corresponding image features of all viewing angles to obtain the identification result, thereby identifying the biological species of the material under test more perfectly and more accurately.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a biological identification method based on material inspection multi-view morphological images and deep learning comprises the following steps:
s1: establishing a model: establishing a convolutional neural network pre-model and pre-training to obtain a proper available convolutional neural network model, and internally arranging a standard biological characteristic database in the convolutional neural network; the establishment of the standard biological characteristic database depends on the existing biological taxonomy knowledge, technicians collect the external morphological image characteristics of each species of organisms according to the existing standard biological classification table, repeatedly input the external morphological image characteristics into the convolutional neural network pre-model in batches according to species classification, train the pre-model to learn the external morphological image characteristics of the corresponding species of organisms from the pre-model, establish a set of judgment rules, store the external morphological image characteristics of the corresponding species of organisms in the convolutional neural network in the form of the standard biological characteristic database, and conveniently make successive analog pairs on sample images in the actual use process to obtain identification results.
S2: material checking and sampling: processing a material to be detected and acquiring external form images of at least two visual angles of the material to be detected; when the external form image of the material to be inspected is collected, at least two visual angles are selected from six standard visual angles such as a top visual angle, a bottom visual angle, a front visual angle, a rear visual angle, a left visual angle, a right visual angle and the like, and the external form of the material to be inspected is shot to obtain the image. For the material to be detected with complex shape, fine structure and composite identification elements, the visual angle is collected in a changeable and changeable way, and a plurality of images are collected, so that the identification accuracy is ensured.
S3: feature extraction: respectively inputting a single external form image, and respectively and correspondingly extracting image features for the single external form image by using a convolutional neural network model;
s4: repeating the step S3 until the image features corresponding to the external form images acquired under different visual angles are extracted;
s5: combining the characteristics: combining each image characteristic to obtain a combined characteristic;
s6: comparison and identification: and comparing the combined features with a standard biological feature database, selecting the species with the minimum feature distance, and outputting the species as an identification result.
Further, S1 includes the following sub-steps:
s11: establishing a convolutional neural network pre-model;
s12: inputting a biological image with known species information into a convolutional neural network pre-model to train the biological image;
s13: and repeating the step S12 for multiple times, so that the convolutional neural network pre-model can learn by itself from the biological image with known species information in the training process, and a biological identification method is established until a proper and available convolutional neural network model is obtained.
Further, S1 further includes: s14: in the convolutional neural network pre-model, a corresponding biological characteristic database is established according to a standard biological retrieval table based on biological taxonomy.
Further, S2 includes the following sub-steps:
s21: cleaning the material to be tested and removing dust and dirt on the surface of the material to be tested;
s22: shooting external form original images of at least two visual angles of a top visual angle, a bottom visual angle and a side visual angle of the material to be detected;
s23: and carrying out white balance processing on the original image, and adjusting the size of the original image to obtain a proper available external form image as a sample image.
Further, S3 includes the following sub-steps:
s31: extracting and cross-layer fusing the high-level features and the bottom-level features of the sample image to obtain embedded features of the sample image;
s32: the embedded features, which are passed forward by the convolutional neural network, are used to obtain second-order information in a compact bilinear pool. Compared with the bottom layer characteristics, the method has richer semantic information, wider visual field and more global analysis angle when the high layer characteristics are additionally extracted from the sample image, so that the method can be used for respectively extracting the bottom layer characteristics and the high layer characteristics from the sample image and enabling the whole convolutional neural network to pay attention to and learn the more subtle, more complex and more diversified image characteristics by using the cross-layer fusion processing method, and has positive significance for improving the identification accuracy of the network.
Further, S31 specifically includes:
s311: a convolutional neural network downsamples an input external form image of a material to be detected to obtain a high-level feature Xq and a plurality of low-level features Xri (i ═ 1.., n);
s312: multiplying the high-level feature with a plurality of low-level feature points to obtain a dot product;
s313: performing probability mapping post-processing on the dot product to obtain a probability graph reflecting the importance degree of each characteristic element;
s314: multiplying the probability map with the corresponding low-level feature points to obtain an attention map;
s315: the attention map is added to the high-level features to obtain the embedded features, Xs, of the input sample image.
In the prior art, when the classification of fine particle degree is realized on multi-channel information, the fine particle degree is generally processed by a method of summing or averaging, and only first-order information can be obtained by the method. In the identification method provided by the invention, Compact bilinear pooling (Compact bilinear Pooling) is adopted to obtain second-order information, and the second-order information is used for distinguishing different categories when the first-order information is the same:
calculating a local descriptor:wherein x is (x)1,x2,...,x|S|,xs∈Rc) For a set of local descriptors, Xs is an embedded feature forwarded by the convolutional neural network;
two sets of local descriptors X and Y are defined, which are compared using a linear kernel machine:
the above formula compares each local descriptor in the first image with each local descriptor in the second image, and the comparison operator is a second-order polynomial kernel, so that the method adopting the two local descriptors compared by the above formula has the discrimination capability of a second-order kernel machine, and helps the convolutional neural network to explain the stronger empirical performance observed in the previous work.
Let k (X, y) denote the comparison kernel, i.e., the second-order polynomial kernel, φ (X) be a low-dimensional projection function, φ (X) e RdWhen d is<<c2, there are<φ(x),φ(y)>K (x, y), i.e., the inner product of the above equation can be approximated as:
the compact bilinear pooling method allows the gradient back propagation of the gradient to realize the end-to-end optimization of the classification pipeline, can obviously compress the characteristics, enables the coding to be more compact, greatly reduces the network and the memory of the classification parameters, and can obtain stronger results.
Further, S3 is optimized by circle loss,
spis internal similarity, snFor inter-class similarity, assuming there are K intra-class scores for x similarity and L inter-class scores for x similarity, thenAndcan be used to represent a score representing both similarities.
Thus, the common circle loss function is expressed as:
wherein [ · therein]+Denotes a truncation at 0, ensureAndare all non-negative numbers. During trainingTime pair of reverse propagationAre respectively multiplied bySuppose thatThe optimum state is Op,The most preferred state is OnWherein (O)n<Op). The score approaches an optimal value when a similarity score moves away from his merit.
Wherein gamma is a scale factor, m set thresholds, deltanSetting a threshold, Δ, for the inter-class distancepSetting a threshold value for the intra-class distance; namely:
the invention has the beneficial effects that: labor is saved: in the technical scheme provided by the invention, manpower is only required to be invested in establishing and pre-training a convolutional neural network pre-model, after the training of the convolutional neural network pre-model is finished, a multi-view external form image of a material to be detected is only required to be acquired through a computer image acquisition system, and the properly available convolutional neural network model obtained through training can be automatically identified and recognized as a sample image, so that the manual operation is reduced to the maximum extent in the whole identification process, the manpower is liberated from tedious, repeated and boring identification work, the dependence of biological species identification work on the manpower and experience is reduced, and the labor cost is saved;
the identification efficiency is high: from the overall view, the convolutional neural network can operate at high speed, process a large number of sample images and obtain corresponding detection results, and compared with a manual identification method, the efficiency is greatly improved;
the accuracy is high: on one hand, in the training process of the convolutional neural network pre-model, a standard biological characteristic database is arranged in the convolutional neural network pre-model to replace original standard data in a traditional deep learning algorithm and provide comparison standards corresponding to biological species for a subsequent actual identification process, on the other hand, along with the continuous improvement of training amount and the continuous improvement of actual identification amount, the convolutional neural network continuously corrects model parameters, continuously learns a judgment method and continuously memorizes image characteristics, and the identification accuracy is continuously improved.
Drawings
Fig. 1 is a flowchart of an insect identification method based on material inspection multi-view morphological images and deep learning, which is implemented in an embodiment.
Fig. 2 is a schematic identification diagram of an insect identification method based on material inspection multi-view morphological images and deep learning, which is implemented in the embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In order to achieve the purpose, the technical scheme of the invention is as follows:
please refer to fig. 1-2.
In one embodiment, the insect identification method based on the material inspection multi-view morphological image and the deep learning includes the following steps:
s1: establishing a model: establishing a convolutional neural network pre-model and pre-training to obtain a proper available convolutional neural network model, and internally arranging a standard insect characteristic database in the convolutional neural network;
s2: material checking and sampling: processing a material to be detected and acquiring external form images of at least two visual angles of the material to be detected; when the external form image of the material to be inspected is collected, at least two visual angles are selected from six standard visual angles such as a top visual angle, a bottom visual angle, a front visual angle, a rear visual angle, a left visual angle, a right visual angle and the like, and the external form of the material to be inspected is shot to obtain the image. For the material to be detected with complex shape, fine structure and composite identification elements, the visual angle is collected in a changeable and changeable way, and a plurality of images are collected, so that the identification accuracy is ensured.
S3: feature extraction: respectively inputting a single external form image, and respectively and correspondingly extracting image features for the single external form image by using a convolutional neural network model;
s4: repeating the step S3 until the image features corresponding to the external form images acquired under different visual angles are extracted;
s5: combining the characteristics: combining each image characteristic to obtain a combined characteristic;
s6: comparison and identification: and comparing the combined characteristics with a standard insect characteristic database, selecting the species with the minimum characteristic distance, and outputting the species as an identification result.
Further, S1 includes the following sub-steps:
s11: establishing a convolutional neural network pre-model;
s12: inputting an insect image with known species information into a convolutional neural network pre-model to train the insect image;
s13: and repeating the step S12 for multiple times, so that the convolutional neural network pre-model can learn by itself from the insect images with known species information in the training process, and an insect identification method is established until a proper and available convolutional neural network model is obtained.
Further, S1 further includes: s14: in the convolutional neural network pre-model, a corresponding insect feature database is established according to a standard insect search table based on insect taxonomy.
Further, S2 includes the following sub-steps:
s21: cleaning the material to be tested and removing dust and dirt on the surface of the material to be tested;
s22: turning on the camera related to device preheating, checking devices, networks, software systems, etc.
The method comprises the steps of preheating the camera equipment and the computer during starting, checking the camera equipment, the computer, the network, the software system and the like, and keeping the hardware equipment, the network, the software system and the like in a normal working state.
S23: shooting a top view, a bottom view, a front view, a rear view, a left view and a right view of a material to be detected;
s24: and performing white balance processing on each view angle view, and adjusting the size of each view angle view to obtain a proper available insect external form image as a sample image.
Further, S3 includes the following sub-steps:
s31: extracting and cross-layer fusing the high-level features and the bottom-level features of the sample image to obtain embedded features of the sample image;
s32: the embedded features, which are passed forward by the convolutional neural network, are used to obtain second-order information in a compact bilinear pool.
Further, S31 specifically includes:
s311: a convolutional neural network downsamples an input external form image of a material to be detected to obtain a high-level feature Xq and a plurality of low-level features Xri (i ═ 1.., n);
s312: multiplying the high-level feature with a plurality of low-level feature points to obtain a dot product;
s313: performing probability mapping on the dot product, and then performing softmax processing to obtain a probability map reflecting the importance degree of each characteristic element;
s314: multiplying the probability map with the corresponding low-level feature points to obtain an attention map of each tiny local part of the insect, wherein the attention map comprises but is not limited to a size attention map, a groove attention map, a glume attention map, a shape attention map, a surface smooth/convex attention map, a pattern attention map, a navel bar attention map and the like;
s315: the attention map is added to the high-level features to obtain the embedded features, Xs, of the input sample image.
In the identification method provided in this embodiment, compact bilinear Pooling (compact bilinear Pooling) is used to obtain second-order information, and when the first-order information is the same, the second-order information is used to distinguish different categories:
calculating a local descriptor:wherein x is (x)1,x2,...,x|S|,xs∈Rc) For a set of local descriptors, Xs is an embedded feature forwarded by the convolutional neural network;
two sets of local descriptors X and Y are defined, which are compared using a linear kernel machine:
the above formula compares each local descriptor in the first image with each local descriptor in the second image, and the comparison operator is a second-order polynomial kernel, so that the method adopting the two local descriptors compared by the above formula has the discrimination capability of a second-order kernel machine, and helps the convolutional neural network to explain the stronger empirical performance observed in the previous work.
Let k (X, y) denote the comparison kernel, i.e., the second-order polynomial kernel, φ (X) be a low-dimensional projection function, φ (X) e RdWhen d is<<c2, there are<(x),φ(y)>K (x, y), i.e., the inner product of the above equation can be approximated as:
the compact bilinear pooling method allows the gradient back propagation of the gradient to realize the end-to-end optimization of the classification pipeline, can obviously compress the characteristics, enables the coding to be more compact, greatly reduces the network and the memory of the classification parameters, and can obtain stronger results.
Further, S3 is optimized by circle loss,
spis internal similarity, snFor inter-class similarity, assuming there are K intra-class scores for x similarity and L inter-class scores for x similarity, thenAndcan be used to represent a score representing both similarities.
Thus, the common circle loss function is expressed as:
wherein [ · therein]+Denotes a truncation at 0, ensureAndare all non-negative numbers. During trainingTime pair of reverse propagationAre respectively multiplied bySuppose thatThe optimum state is Op,The most preferred state is OnWherein (O)n<Op). The score approaches an optimal value when a similarity score moves away from his merit.
wherein gamma is a scale factor, m set thresholds, deltanSetting a threshold, Δ, for the inter-class distancepSetting a threshold value for the intra-class distance; namely:
the present invention is not limited to the above preferred embodiments, and any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. A biological identification method based on material inspection multi-view morphological images and deep learning is characterized by comprising the following steps:
s1: establishing a model: establishing a convolutional neural network pre-model and pre-training to obtain a proper available convolutional neural network model, and internally arranging a standard biological characteristic database in the convolutional neural network;
s2: material checking and sampling: processing a material to be detected and acquiring external form images of at least two visual angles of the material to be detected;
s3: feature extraction: respectively inputting a single external form image, and respectively and correspondingly extracting image features for the single external form image by using a convolutional neural network model;
s4: repeating the step S3 until the image features corresponding to the external form images acquired under different visual angles are extracted;
s5: combining the characteristics: combining each image characteristic to obtain a combined characteristic;
s6: comparison and identification: and comparing the combined features with a standard biological feature database, selecting the species with the minimum feature distance, and outputting the species as an identification result.
2. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning as claimed in claim 1, wherein the S1 comprises the following sub-steps:
s11: establishing a convolutional neural network pre-model;
s12: inputting a biological image with known species information into a convolutional neural network pre-model to train the biological image;
s13: and repeating the step S12 for multiple times, so that the convolutional neural network pre-model can learn by itself from the biological image with known species information in the training process, and a biological identification method is established until a proper and available convolutional neural network model is obtained.
3. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning of claim 2, wherein the S1 further comprises: s14: in the convolutional neural network pre-model, a corresponding biological characteristic database is established according to a standard biological retrieval table based on biological taxonomy.
4. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning as claimed in claim 1, wherein the S2 comprises the following sub-steps:
s21: cleaning the material to be tested and removing dust and dirt on the surface of the material to be tested;
s22: shooting external form original images of at least two visual angles of a top visual angle, a bottom visual angle and a side visual angle of the material to be detected;
s23: and carrying out white balance processing on the original image, and adjusting the size of the original image to obtain a proper available external form image as a sample image.
5. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning as claimed in claim 1, wherein the S3 comprises the following sub-steps:
s31: extracting and cross-layer fusing the high-level features and the bottom-level features of the sample image to obtain embedded features of the sample image;
s32: the embedded features, which are passed forward by the convolutional neural network, are used to obtain second-order information in a compact bilinear pool.
6. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning as claimed in claim 5, wherein the step S31 is specifically as follows:
s311: a convolutional neural network downsamples an input external form image of a material to be detected to obtain a high-level feature Xq and a plurality of low-level features Xri (i ═ 1.., n);
s312: multiplying the high-level feature with a plurality of low-level feature points to obtain a dot product;
s313: performing probability mapping post-processing on the dot product to obtain a probability graph reflecting the importance degree of each characteristic element;
s314: multiplying the probability map with the corresponding low-level feature points to obtain an attention map;
s315: the attention map is added to the high-level features to obtain the embedded features, Xs, of the input sample image.
7. The biometric authentication method based on the material inspection multi-view morphological image and the deep learning as claimed in claim 6, wherein the step S32 is specifically as follows:
calculating a local descriptor:wherein x is (x)1,x2,...,x|S|,xs∈Rc) For a set of local descriptors, Xs is an embedded feature forwarded by the convolutional neural network;
two sets of local descriptors X and Y are defined, which are compared using a linear kernel machine:
when d < < c2, there are:
8. the biometric authentication method based on the inspection material multi-view morphological image and the deep learning as claimed in claim 1, wherein the S3 is optimized by circle loss, specifically:
wherein gamma is a scale factor, m set thresholds, deltanSetting a threshold, Δ, for the inter-class distancepSetting a threshold value for the intra-class distance; namely:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010759848.8A CN111898680A (en) | 2020-07-31 | 2020-07-31 | Biological identification method based on material inspection multi-view morphological image and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010759848.8A CN111898680A (en) | 2020-07-31 | 2020-07-31 | Biological identification method based on material inspection multi-view morphological image and deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111898680A true CN111898680A (en) | 2020-11-06 |
Family
ID=73183930
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010759848.8A Pending CN111898680A (en) | 2020-07-31 | 2020-07-31 | Biological identification method based on material inspection multi-view morphological image and deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111898680A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116704356A (en) * | 2023-08-08 | 2023-09-05 | 吉林省农业科学院 | Corn salt and alkali tolerance identification method and system based on convolutional neural network |
-
2020
- 2020-07-31 CN CN202010759848.8A patent/CN111898680A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116704356A (en) * | 2023-08-08 | 2023-09-05 | 吉林省农业科学院 | Corn salt and alkali tolerance identification method and system based on convolutional neural network |
CN116704356B (en) * | 2023-08-08 | 2023-10-03 | 吉林省农业科学院 | Corn salt and alkali tolerance identification method and system based on convolutional neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108171184B (en) | Method for re-identifying pedestrians based on Simese network | |
CN108960140B (en) | Pedestrian re-identification method based on multi-region feature extraction and fusion | |
CN111709311B (en) | Pedestrian re-identification method based on multi-scale convolution feature fusion | |
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
CN103279768B (en) | A kind of video face identification method based on incremental learning face piecemeal visual characteristic | |
CN112633382B (en) | Method and system for classifying few sample images based on mutual neighbor | |
CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
CN111178197A (en) | Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method | |
CN106951551B (en) | Multi-index image retrieval method combining GIST characteristics | |
Rasyidi et al. | Batik pattern recognition using convolutional neural network | |
Termritthikun et al. | NU-InNet: Thai food image recognition using convolutional neural networks on smartphone | |
CN107992783A (en) | Face image processing process and device | |
CN112766229A (en) | Human face point cloud image intelligent identification system and method based on attention mechanism | |
CN110580510A (en) | clustering result evaluation method and system | |
Sun et al. | Brushstroke based sparse hybrid convolutional neural networks for author classification of Chinese ink-wash paintings | |
CN111898680A (en) | Biological identification method based on material inspection multi-view morphological image and deep learning | |
CN110083724A (en) | A kind of method for retrieving similar images, apparatus and system | |
CN110909678B (en) | Face recognition method and system based on width learning network feature extraction | |
CN110738194A (en) | three-dimensional object identification method based on point cloud ordered coding | |
Du et al. | Image retrieval based on colour and improved NMI texture features | |
CN115795394A (en) | Biological feature fusion identity recognition method for hierarchical multi-modal and advanced incremental learning | |
CN115100694A (en) | Fingerprint quick retrieval method based on self-supervision neural network | |
CN114627496A (en) | Robust pedestrian re-identification method based on depolarization batch normalization of Gaussian process | |
CN110348323B (en) | Wearable device gesture recognition method based on neural network optimization | |
CN113887509A (en) | Rapid multi-modal video face recognition method based on image set |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |