RTMDet-based oil palm tree identification method and system, medium and electronic equipment

Through the RTMDet-based oil palm tree recognition method, feature extraction and fusion is used using the backbone network and neck module, combined with SimSPPF and SGEAttention technology, the problem of low counting efficiency of traditional oil palm tree is solved, and the precise detection of oil palm tree targets is achieved and quick identification.

CN118918409APending Publication Date: 2024-11-08GUILIN UNIVERSITY OF TECHNOLOGY
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410743371.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-11
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The traditional oil palm tree counting method relies on manual statistics, has low efficiency and poor results, and the existing remote sensing technology has not been effectively applied in oil palm tree target detection.

Method used

Using the RTMDet-based oil palm tree recognition method, the prediction model is trained by obtaining and labeling the oil palm tree data set, and feature extraction and fusion is used to achieve accurate detection of oil palm tree targets. Technologies such as SimSPPF, CSP-ELANLayer and SGEAttention improve the adaptability and operation speed of the model.

Benefits of technology

The precise detection and rapid identification of oil palm tree targets are achieved, the adaptability and operation speed of the model in complex situations is improved, and the problem of low efficiency of traditional manual counting is solved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918409A_ABST
    Figure CN118918409A_ABST
Patent Text Reader

Abstract

The invention provides an oil palm tree identification method based on RTMDet, and the method is characterized in that the method comprises the steps: obtaining an oil palm tree data set; the oil palm tree data set is used for training, feature information of the oil palm tree is learned, a trained prediction model is obtained, and the prediction model is obtained based on RTMDet; and inputting an oil palm tree image needing to be predicted into the trained prediction model, and detecting the oil palm tree. According to the method, an SimSPPF structure is used for replacing an SPPF structure, a CSP-ELANLayer structure is introduced, UpSamplingbiliearn is used as an up-sampling mode, and finally an SGEAttention mechanism is introduced, so that the extraction capability of key features of the oil palm tree is improved, accurate detection of an oil palm tree target can be realized, the adaptability of the model is relatively high under different complex conditions, and meanwhile, the operation speed of the model is relatively high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of target detection, and in particular relates to an oil palm tree recognition method, system, medium and electronic equipment based on RTMDet. Background Art

[0002] As an important cash crop, oil palm is often used as a raw material for vegetable oil, food, cosmetics, detergents and other products. At the same time, the oil production of oil palm is 6 times that of other oilseeds. It is an important oil crop and occupies an important position in the world oil market. At present, oil palm plantations are mostly distributed in Southeast Asia, Africa and Latin America, mainly concentrated in tropical countries such as Indonesia, Malaysia and Thailand. Oil palm tree counting is the basis for oil palm plantation yield estimation, plantation management and spatial layout planning.

[0003] Traditionally, counting of oil palm trees is done manually, which is time-consuming, labor-intensive and ineffective. Remote sensing technology, especially high-resolution remote sensing images, provides a new way to count oil palm trees. Based on high-resolution remote sensing images, oil palm tree target detection can be performed through image processing and deep learning algorithms. Summary of the invention

[0004] The object of the present invention is to provide an oil palm tree identification method, system, medium and electronic device based on RTMDet.

[0005] The technical solution of the present invention is as follows:

[0006] A method for identifying an oil palm tree based on RTMDet, the method comprising:

[0007] Get the oil palm tree dataset;

[0008] Using the oil palm tree dataset for training, learning characteristic information of the oil palm tree, and obtaining a trained prediction model, wherein the prediction model is obtained based on RTMDet;

[0009] The oil palm tree image to be predicted is input into the trained prediction model to detect the oil palm tree.

[0010] Furthermore, the oil palm tree dataset is used for training to learn the characteristic information of the oil palm tree, and the trained prediction model is specifically:

[0011] Get the oil palm tree dataset;

[0012] Input the oil palm tree dataset into the backbone network to extract feature maps;

[0013] The extracted feature map is input into the neck module for feature layer fusion;

[0014] Input the fused feature layer into the detection head for target detection training to obtain the optimal model weights;

[0015] A trained prediction model is obtained according to the optimal model weights.

[0016] Furthermore, the step of obtaining the oil palm tree dataset is as follows:

[0017] Collect images of oil palm trees;

[0018] The oil palm trees in the oil palm tree image are marked to create an oil palm tree dataset.

[0019] The present invention also provides an oil palm tree identification system based on RTMDet, the system comprising:

[0020] A data acquisition module, used to obtain the oil palm tree dataset;

[0021] A model training module, used for training with the oil palm tree dataset, learning characteristic information of the oil palm tree, and obtaining a trained prediction model, wherein the prediction model is obtained based on RTMDet;

[0022] The prediction module is used to input the oil palm tree image to be predicted into the trained prediction model to detect the oil palm tree.

[0023] Further, the prediction model is composed of a backbone network, a neck, and a detection head;

[0024] The backbone network is used to extract oil palm tree feature information;

[0025] The neck is used for feature fusion;

[0026] The detection head is used to predict the location and category of objects in the image.

[0027] Further, the backbone network is composed of a first CBS module, CSPBlock and SimSPPF;

[0028] The first CBS module is used to establish a connection between the deep feature map and the shallow feature map;

[0029] The CSPBlock is used to strengthen the receptive field of the model;

[0030] The SimSPPF is used to process information of different scales.

[0031] Further, the neck is composed of CSP-ELANLayer, a second CBS module, UpsampleBiliearn and SGEAttention;

[0032] The CSP-ELANLayer is used to process feature fusion and integrate features from different levels;

[0033] The second CBS module is used to guide the gradient flow to different branches;

[0034] The UpsampleBiliearn is used for upsampling so that the size of the feature map matches the original image;

[0035] The SGE Attention is used to enhance the feature extraction capability and input the fused feature layer into the detection head.

[0036] The present invention also provides an electronic device, characterized in that it includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned RTMDet-based oil palm tree identification method.

[0037] The present invention also provides a computer-readable storage medium, characterized in that it stores a computer program, and the computer program is executed by a processor to implement the above-mentioned RTMDet-based oil palm tree identification method.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] In the present invention, SimSPPF is used to replace the SPPF structure, the CSP-ELANLayer structure is introduced, UpSamplingbiliearn is used as the upsampling method, and finally the SGE Attention mechanism is introduced to improve the extraction ability of key features of oil palm trees, so that accurate detection of oil palm tree targets can be achieved, and the model has strong adaptability in different complex situations, and the model runs faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0041] Figure 1 A schematic diagram of the improved prediction model of the present invention is shown;

[0042] Figure 2 The SPPF structure and SimSPPF structure schematic diagram of the present invention are shown;

[0043] Figure 3The schematic diagram of the structure of CSPNeXtBlock, CSPLayer and CSP-ELANLayer of the present invention is shown;

[0044] Figure 4 The schematic diagram of the SGE Attention structure of the present invention is shown;

[0045] Figure 5 A schematic diagram of the process of the present invention is shown. DETAILED DESCRIPTION

[0046] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] The embodiment of the present invention provides an oil palm tree identification method based on RTMDet, which specifically includes

[0048] Collecting oil palm tree images, marking the oil palm trees in the oil palm tree images, and creating an oil palm tree dataset;

[0049] Using the oil palm tree dataset for training, learning characteristic information of the oil palm tree, and obtaining a trained prediction model, wherein the prediction model is obtained based on RTMDet;

[0050] The oil palm tree image to be predicted is input into the trained prediction model to detect the oil palm tree.

[0051] The oil palm tree dataset is used for training to learn the characteristic information of the oil palm tree, and the trained prediction model is specifically:

[0052] Get the oil palm tree dataset;

[0053] Input the oil palm tree dataset into the backbone network to extract feature maps;

[0054] The extracted feature map is input into the neck module for feature layer fusion;

[0055] Input the fused feature layer into the detection head for target detection training to obtain the optimal model weights;

[0056] A trained prediction model is obtained according to the optimal model weights.

[0057] As another embodiment of an oil palm tree identification method based on RTMDet, the invention comprises:

[0058] The present invention mainly comprises three parts: backbone, neck and head. Backbone is composed of CSPNet network, which is used to extract oil palm tree feature information. Neck is a feature pyramid structure composed of top-down paths and bottom-up paths, which enhances feature expression ability and information flow, thereby significantly improving performance. Head is composed of a detection head with shared weights. The detection head parameter weights used by the three feature layers are shared, but the head does not use batch normalization operation. Such a structure reduces the number of parameters of the head while maintaining the accuracy of the model.

[0059] As a single-stage object detector, the RTMDet model is highly versatile and can be used for tasks such as instance segmentation and remote sensing image segmentation. Oilplam-RTMDet is designed based on the RTMDet model to detect oil palm faster and more accurately in complex situations. The workflow of the Oilplam-RTMDet model is as follows:

[0060] At the beginning of the training process, the image data to be trained is input into the OilPlam-RTMDet model. Inside the model, after Mixup and Mosaic processing with an improved caching mechanism, the image enters the training pipeline. These technologies can effectively increase the diversity of data and improve the generalization ability of the model.

[0061] The model consists of several key modules, including Backbone, Neck and Head. Backbone mainly consists of the first CBS module, CSPBlock and SimSPPF, which are used to extract image features. The first CBS module is used to establish the connection between the deep feature map and the shallow feature map, CSPLayer is used to enhance the receptive field of the model, and SimSPPF is used to process information of different scales.

[0062] The Neck part is mainly composed of a feature pyramid structure consisting of CSP-ELANLayer, the second CBS module, UpsampleBiliearn and SGEAttention. CSP-ELANLayer is used to process feature fusion and integrate features from different levels; the second CBS module is used to guide the gradient flow to different branches; UpsampleBiliearn is used for upsampling so that the size of the feature map matches the original image. In addition, the PAFPN structure is adopted so that features from different scales can be fully integrated, thereby improving the detection performance of the model. It is input into the Head through SGEAttention.

[0063] Finally, after the Neck part processes the features, the three fused feature layers are input into the weight-sharing detection head for object detection. This detection head predicts the location and category of the object in the image based on the input feature map. Through this process, the model can effectively detect the object in the image and generate the corresponding detection results.

[0064] The OilPlam-RTMDet model designed by the present invention is divided into two stages: offline training and online detection. The offline training stage refers to the use of the marked training oil palm tree data for training before the model is officially deployed online, learning the characteristic information of the oil palm tree and obtaining the weight of the model. The online detection stage refers to inputting the oil palm tree image to be predicted into the model, and detecting the oil palm tree using the best training model weight obtained by training. In the present invention, SimSPPF is used to replace the SPPF structure, the CSP-ELANLayer structure is introduced in the feature fusion module, and UpSamplingbiliearn is used as the upsampling method. Finally, the SGEAttention mechanism is introduced to improve the extraction ability of the key features of the oil palm tree, so that the accurate detection of the oil palm tree target can be achieved, and the adaptability of the model under different complex conditions is strong, and the model runs faster.

[0065] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for identifying oil palm trees based on RTMDet, It is characterized in that The method comprises: Get the oil palm tree dataset; Using the oil palm tree dataset for training, learning characteristic information of the oil palm tree, and obtaining a trained prediction model, wherein the prediction model is obtained based on RTMDet; The oil palm tree image to be predicted is input into the trained prediction model to detect the oil palm tree.

2. the oil palm identification method based on RTMDet according to claim 1, It is characterized in that The oil palm tree dataset is used for training to learn the characteristic information of the oil palm tree, and the trained prediction model is specifically: Get the oil palm tree dataset; Input the oil palm tree dataset into the backbone network to extract feature maps; The extracted feature map is input into the neck module for feature layer fusion; Input the fused feature layer into the detection head for target detection training to obtain the optimal model weights; A trained prediction model is obtained according to the optimal model weights.

3. the oil palm identification method based on RTMDet according to claim 1, It is characterized in that The specific steps of obtaining the oil palm tree dataset are as follows: Collect images of oil palm trees; The oil palm trees in the oil palm tree image are marked to create an oil palm tree dataset.

4. An oil palm tree identification system based on RTMDet, It is characterized in that The system comprises: A data acquisition module, used to obtain the oil palm tree dataset; A model training module, used for training with the oil palm tree dataset, learning characteristic information of the oil palm tree, and obtaining a trained prediction model, wherein the prediction model is obtained based on RTMDet; The prediction module is used to input the oil palm tree image to be predicted into the trained prediction model to detect the oil palm tree.

5. the oil palm identification method based on RTMDet according to claim 4, It is characterized in that The prediction model consists of a backbone network, a neck and a detection head; The backbone network is used to extract oil palm tree feature information; The neck is used for feature fusion; The detection head is used to predict the location and category of objects in the image.

6. the oil palm identification method based on RTMDet according to claim 5, It is characterized in that The backbone network consists of a first CBS module, CSPBlock and SimSPPF; The first CBS module is used to establish a connection between the deep feature map and the shallow feature map; The CSPBlock is used to strengthen the receptive field of the model; The SimSPPF is used to process information of different scales.

7. the oil palm identification method based on RTMDet according to claim 5, It is characterized in that The neck consists of CSP-ELANLayer, the second CBS module, UpsampleBiliearn and SGEAttention; The CSP-ELANLayer is used to process feature fusion and integrate features from different levels; The second CBS module is used to guide the gradient flow to different branches; The UpsampleBiliearn is used for upsampling so that the size of the feature map matches the original image; The SGE Attention is used to input the fused feature layer into the detection head.

8. An electronic device, It is characterized in that It comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the oil palm tree identification method based on RTMDet according to any one of claims 1 to 3.

9. A computer-readable storage medium, It is characterized in that It stores a computer program, which, when executed by a processor, implements the oil palm tree identification method based on RTMDet as described in any one of claims 1 to 3.

Citation Information

Patent Citations

  • Detection method based on small-scale pine wood nematode disease tree

    CN115063700A

  • Two-dimensional hand posture estimation method and device, equipment and storage medium

    CN115511968A

  • Natural tree species identification method based on improved YOLOv7

    CN115578602A

  • Target detection and classification method and device based on optical remote sensing image

    CN117475320A