A crop classification method based on 3DUNet
Through the 3DUNet-based crop classification method, 3D image encoding and decoding technology, combined with database comparison, the refined classification of crops and the rapid identification of multiple crops is achieved, and the general and time-consuming classification in the existing technology is solved. .
Patent Information
- Application Number
- CN202310882040.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-18
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-07-18
AI Technical Summary
In the prior art, when classifying crops, classification methods are mostly general and inconvenient to refinement, and can only identify one crop at a time, resulting in a lot of time being spent in classification of multiple crops.
Using the crop classification method based on 3DUNet, the crops are initially divided into four parts and subdivided. After establishing a database, 3D images are collected for the classified crops, input 3DUNet for encoding, dimensionality reduction and reverse decoding, and generate segmented images and Database comparison to achieve refined classification and simultaneous classification of multiple crops.
The refined classification of crops is realized, the classification efficiency and speed are improved, and multiple crops can be classified at the same time, solving the problem that most classifications in the prior art are general and time-consuming.
Smart Images

Figure CN116740468B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of crop classification, and in particular to a crop classification method based on 3DUNet. Background Art
[0002] 3D UNet is a deep learning model used in the field of medical image segmentation. The basic principle is to input a 3D image into an encoder and reduce its dimension through a series of convolutional layers. The output of the encoder is then upsampled through a series of deconvolutional layers, and the final output is a segmentation result with the same size as the original input. Related questions: 3D unet uses matlab to segment MRI 3D images MRI 3D image segmentation is an important task in the field of medical image analysis, which can help doctors better diagnose the disease. 3D unet is a deep learning network structure that can be used for image segmentation tasks. It contains an encoder and a decoder. The encoder gradually compresses the input image into a smaller feature vector, and the decoder reverses the feature vector and outputs the segmented image. At the same time, 3D UNet can also be used for crop classification.
[0003] For example, the Chinese patent with publication number CN109409261B discloses a crop classification method and system. The classification method includes: constructing historical crop labeling samples based on historical multi-phase medium-resolution images and historical crop phenological calendar data; constructing high-precision small-area current crop labeling samples based on current multi-phase high-resolution images and current crop phenological calendar data; establishing a remote sensing network RSNet model based on the ResNet model and the PSPNet model; pre-training the remote sensing network RSNet model based on historical training samples to establish a pre-trained remote sensing network RSNet model; fine-tuning the pre-trained remote sensing network RSNet model based on current training samples to establish a fine-tuned remote sensing network RSNet model; establishing an optimal pre-trained remote sensing network RSNet model and an optimal fine-tuned remote sensing network RSNet model; classifying the crops based on the optimal pre-trained remote sensing network RSNet model and the optimal fine-tuned remote sensing network RSNet model. The classification method and system provided by the present invention can reduce the workload of selecting training samples and improve classification efficiency.
[0004] The existing technology has the following problems:
[0005] When the existing technology classifies crops, the existing classification methods are mostly remote sensing classification, but this classification method is not convenient to use when identifying and classifying crop images; when the existing technology classifies crops, the classification is mostly general, which is not convenient for fine classification of crops and has certain inconveniences in use; when the existing technology classifies crops, only one crop can be identified and classified at a time. When there are many types to be classified, it takes a lot of time and is not convenient to use. Summary of the invention
[0006] The present invention provides a crop classification method based on 3DUNet to solve the problems raised in the above background technology.
[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0008] A crop classification method based on 3DUNet comprises the following steps;
[0009] Step 1: Database establishment;
[0010] Step 2: Image acquisition;
[0011] Step 3: Image compression;
[0012] Step 4: Image decoding;
[0013] Step 5: Database comparison;
[0014] Step 6: Crop classification.
[0015] A further improvement of the technical solution of the present invention is that: the step one includes: dividing the crops into four major parts: food crops, cash crops, fodder and green manure crops, and medicinal crops.
[0016] A further improvement of the technical solution of the present invention is that the step one comprises: after the crops are divided into four major parts, the four major parts are further subdivided, food crops are divided into: cereal crops, legume crops, taro and tuber crops, economic crops are divided into: fiber crops, oil crops, sugar crops, and other crops, feed and green manure crops are divided into feed crops and green manure crops, and medicinal crops are divided into common ginseng, wolfberry, adenophora, etc. After the classification is completed, images of the growth appearance of various crops in each period are collected, and after the collection, they are input into the database according to the classification to establish the database.
[0017] A further improvement of the technical solution of the present invention is that: the step 2 includes: after the database is established, 3D images of the crops to be classified are collected, and after the crop 3D images are collected, they are input into the encoder in 3DUNet for dimensionality reduction.
[0018] A further improvement of the technical solution of the present invention is that: the step three includes: the encoder in 3DUNet compresses the collected 3D image into a smaller feature orientation, and then the encoder transmits the sub-feature orientation to the decoder in 3DUNet for model decoding.
[0019] A further improvement of the technical solution of the present invention is that: the step four includes: after receiving the directional feature transmitted from the encoder, the decoder of 3DUNet reversely decodes the directional feature and outputs multiple segmented images of the collected crop 3D image.
[0020] A further improvement of the technical solution of the present invention is that the step five includes: by comparing the multiple segmented images output by the decoder with the images of each type of crop in the database in turn, the classification of the crop can be obtained.
[0021] A further improvement of the technical solution of the present invention is that step six includes: after comparing the segmented image with the database, selecting a type of crop with the highest similarity to complete crop classification, and by comparing multiple segmented images, multiple crops in the same image can be accurately classified at the same time.
[0022] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:
[0023] 1. The present invention provides a crop classification method based on 3DUNet, which collects 3D images of crops to be classified, and then inputs the collected crop 3D images into an encoder in 3DUNet for encoding and dimension reduction. Under the action of the encoder in 3DUNet, the image is compressed into a feature orientation, and then the feature orientation is input into a decoder in 3DUNet, and the feature orientation is reversely operated to output a segmented image. After the segmented image is obtained, the segmented image is compared with the data in the database, and the option with the highest similarity is selected to classify the crops, thereby solving the problem that most existing classification methods are remote sensing classification, which is not convenient to use when identifying and classifying crop images.
[0024] 2. The present invention provides a crop classification method based on 3DUNet. Through the division in the early stage when establishing a database, crops are initially divided into four major parts: food crops, economic crops, fodder and green manure crops, and medicinal crops. The crops are first generally classified, and then the crops are subdivided again. Among them, food crops are further subdivided into cereal crops, legume crops, and taro and tuber crops, and economic crops are further subdivided into fiber crops, oil crops, sugar crops and other crops, so that they are divided into four major parts and nine categories in total, so that the classification of crops is more refined and more convenient to use. After the classification, the growth period images of various crops are uploaded to the database to fill the database. After the database is filled, the crops can be identified and classified, which solves the problem that when classifying crops, the classification is mostly relatively general, which is not convenient for fine classification of crops and has certain inconveniences in use.
[0025] 3. The present invention provides a crop classification method based on 3DUNet. 3DUNet can be used to segment multiple crops in an image, so that multiple crops can be classified at the same time during crop classification. The classification speed is faster and easier to use. This solves the problem that when classifying crops, only one crop can be identified and classified at a time. When there are many types to be classified, it takes a lot of time and is inconvenient to use. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION
[0027] The present invention is further described in detail below in conjunction with embodiments:
[0028] Example 1
[0029] like Figure 1 As shown, the present invention provides a crop classification method based on 3DUNet, comprising the following steps:
[0030] Step 1: Database establishment;
[0031] Step 2: Image acquisition;
[0032] Step 3: Image compression;
[0033] Step 4: Image decoding;
[0034] Step 5: Database comparison;
[0035] Step 6: Crop classification.
[0036] Step one includes: dividing crops into four major parts: food crops, cash crops, fodder and green manure crops, and medicinal crops. After the crops are divided into four major parts, the four major parts are further subdivided. Food crops are divided into: cereal crops, legume crops, taro and tuber crops. Cash crops are divided into: fiber crops, oil crops, sugar crops, and other crops. Fodder and green manure crops are divided into fodder crops and green manure crops. Medicinal crops are divided into common ginseng, wolfberry, adenophora, etc. After the classification is completed, images of the growth appearance of various crops in each period are collected, and after the collection, they are input into the database according to the classification to establish the database.
[0037] In this embodiment, through the early division when establishing the database, crops are initially divided into four major parts: food crops, cash crops, fodder and green manure crops, and medicinal crops. The crops are first generally classified, and then the crops are subdivided again. Among them, food crops are further subdivided into cereal crops, legume crops, and taro crops, and cash crops are further subdivided into fiber crops, oil crops, sugar crops and other crops, so that they are divided into four major parts and nine categories in total, making the classification of crops more refined and easier to use. After the classification, the growth period images of various crops are uploaded to the database to fill the database. After the database is filled, the crops can be identified and classified, which solves the problem that when classifying crops, the classification is mostly general, which is not convenient for fine classification of crops and has certain inconveniences in use.
[0038] Example 2
[0039] like Figure 1 As shown, on the basis of Example 1, the present invention provides a technical solution: preferably, step 2 includes: after the database is established, 3D images of crops to be classified are collected, and after the crop 3D images are collected, they are input into the encoder in 3DUNet for dimensionality reduction, and step 3 includes: the encoder in 3DUNet compresses the collected 3D image into a smaller feature orientation, and then the encoder transmits the sub-feature orientation to the decoder in 3DUNet for model decoding. After receiving the directional feature transmitted from the encoder, the decoder of 3DUNet reversely decodes the directional feature and outputs multiple segmented images of the collected crop 3D image.
[0040] In this embodiment, when performing classification, 3D images of crops to be classified are first collected. After the collection is completed, the collected crop 3D images are input into the encoder in 3DUNet for encoding and dimensionality reduction. Under the action of the encoder in 3DUNet, the image is compressed into a feature orientation, and then the feature orientation is input into the decoder in 3DUNet. Under the action of the decoder, the feature orientation is reversed to output a segmented image. After the segmented image is obtained, the segmented image is compared with the data in the database. When comparing, the option with the highest similarity is selected to classify the crops. This solves the problem that when classifying crops, the existing classification methods are mostly remote sensing classification, but this classification method is not convenient to use when identifying and classifying crop images.
[0041] Example 3
[0042] like Figure 1 As shown, on the basis of Example 1, the present invention provides a technical solution: preferably, by comparing the multiple segmented images output by the decoder with the images of each type of crop in the database in turn, the classification of the crops can be known, and after comparing the segmented images with the database, the type of crops with the highest similarity is selected to complete the crop classification, and by comparing multiple segmented images, multiple crops in the same image can be accurately classified at the same time.
[0043] In this embodiment, 3DUNet can be used to segment multiple crops in an image, so that multiple crops can be classified at the same time during crop classification, with faster classification speed and easier use. This solves the problem that when classifying crops, only one crop can be identified and classified at a time, which takes a lot of time and is inconvenient to use when there are many types to be classified.
[0044] In summary, when the present invention is used, the crops are initially divided into four major parts through the division in the early stage when the database is established: food crops, economic crops, fodder and green manure crops, and medicinal crops. The crops are first generally classified, and then the crops are subdivided again. Among them, food crops are further subdivided into cereal crops, legume crops, and taro crops, and economic crops are further subdivided into fiber crops, oil crops, sugar crops and other crops, so that they are divided into four major parts and nine categories in total, making the classification of crops more refined and easier to use. After the classification, the growth period images of various crops are uploaded to the database to fill the database. After the database is filled, the crops can be identified and classified. When classifying, first treat the classified crops as 3D images of crops are collected. After the collection is completed, the collected crop 3D images are input into the encoder in 3DUNet for encoding and dimensionality reduction. Under the action of the encoder in 3DUNet, the image is compressed into a feature orientation, and then the feature orientation is input into the decoder in 3DUNet. Under the action of the decoder, the feature orientation is reversed to output a segmented image. After the segmented image is obtained, the segmented image is compared with the data in the database. When comparing, the option with the highest similarity is selected to classify the crops. At the same time, 3DUNet can segment multiple crops in the image, so that multiple crops can be classified at the same time when classifying crops, which is faster and more convenient to use.
[0045] The above generally describes the present invention in detail, but it is obvious to a person skilled in the art that some modifications or improvements can be made to the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.
Claims
1. A crop classification method based on 3DUNet, characterized by: The steps include: Step 1: Database establishment; divide crops into four major parts: food crops, cash crops, fodder and green manure crops, and medicinal crops; after dividing crops into four major parts, further subdivide the four major parts, food crops are divided into: cereal crops, legume crops, taro and tuber crops, cash crops are divided into: fiber crops, oil crops, sugar crops, and other crops, fodder and green manure crops are divided into fodder crops and green manure crops, after the classification is completed, collect images of the growth appearance of various crops in each cycle, and enter the images into the database according to the classification to establish the database; Through the division in the early stage of establishing the database, crops are initially divided into four major parts: food crops, cash crops, fodder and green manure crops, and medicinal crops. Crops are first generally classified, and then the crops are further subdivided. Among them, food crops are further subdivided into cereal crops, legume crops, and taro crops, and cash crops are further subdivided into fiber crops, oil crops, sugar crops and other crops, so that they are divided into four major parts and nine categories, making the classification of crops more refined and easier to use. After the classification, the growth period images of various crops are uploaded to the database to fill the database; Step 2: Image acquisition: After the database is established, 3D images of the crops to be classified are collected. After the crop 3D images are collected, they are input into the encoder in 3DUNet for dimensionality reduction. Step 3: Image compression: The encoder in 3DUNet compresses the collected 3D image into a smaller feature orientation, and then the encoder transmits the feature orientation to the decoder in 3DUNet for model decoding; Step 4: Image decoding: After receiving the directional features transmitted from the encoder, the decoder of 3DUNet reversely decodes the directional features and outputs multiple segmented images of the collected crop 3D image; Step 5: Database comparison: By comparing the multiple segmented images output by the decoder with the images of each type of crop in the database, the classification of the crop can be obtained; Step 6: Crop classification; after comparing the segmented image with the database, select the crop with the highest similarity to complete the crop classification. At the same time, by comparing multiple segmented images, multiple crops in the same image can be accurately classified at the same time.
