Method and system for detecting plant root pores in soil CT image based on improved SegNet model

Abstract

The invention discloses a method and system for detecting plant root pores in a soil CT image based on an improved SegNet model, and belongs to the technical field of computer vision. The method comprises the following steps: acquiring a soil CT image set, and preprocessing the soil CT image set to obtain a target CT image set; the SegNet network model is improved, and an improved SegNet network model is obtained; dividing the target CT image set into a training set, a verification set and a test set according to a preset proportion, and training the improved SegNet network model according to the training set and the verification set; and inputting the test set into the trained improved SegNet network model, outputting the position of the plant root pore in the test set, and completing the detection of the plant root pore in the soil CT image. By improving the SegNet model, the plant root porosity in the soil CT image can be identified more accurately, false detection and missing detection conditions are reduced, and the detection efficiency and the detection precision are improved.

Description

Technical Field

[0001] This application belongs to the field of computer vision, and particularly relates to a method and system for detecting plant root pores in soil CT images based on an improved SegNet model. Background Art

[0002] In agricultural science and soil science research, accurately evaluating the distribution of plant roots in soil and their pore structures is crucial for understanding the interactions between soil-plant, pore-water, and water-root, optimizing soil management, and increasing crop yields. Traditional soil root research methods, such as excavation and root washing, are not only time-consuming and laborious but may also damage the root structure, thus affecting the accuracy of the results. In recent years, with the development of computer tomography (CT) technology, soil CT images have become a research direction for plant roots and their pores; detecting plant root pores in soil CT images through machine learning has also gradually become a trend.

[0003] Currently, the detection of plant root pores in artificial soil CT images based on visual recognition has low detection efficiency, poor detection accuracy, and is particularly difficult to cut for complex pore networks in the field, severely restricting the research in related fields and the development of science and technology. Summary of the Invention

[0004] The purpose of the embodiments of this application is to provide a method and system for detecting plant root pores in soil CT images based on an improved SegNet model, which can solve the problems of low efficiency, low accuracy, and difficult cutting in accurately identifying plant root pores from complex soil CT images in the prior art.

[0005] To solve the above technical problems, this application is implemented as follows:

[0006] In a first aspect, the embodiments of this application provide a method for detecting plant root pores in soil CT images based on an improved SegNet model, and the method includes:

[0007] Obtain a soil CT image set, preprocess the soil CT image set to obtain a target CT image set;

[0008] Improve the SegNet network model to obtain an improved SegNet network model;

[0009] Divide the target CT image set into a training set, a validation set, and a test set according to a preset ratio, and train the improved SegNet network model according to the training set and the validation set;

[0010] Input the test set into the trained improved SegNet network model, and output the positions of plant root pores in the test set to complete the detection of plant root pores in soil CT images.

[0011] As an optional implementation manner of the first aspect of this application, the method for obtaining a soil CT image set and preprocessing the soil CT image set to obtain a target CT image set is as follows:

[0012] Obtain soil samples from multiple regions, and divide the soil samples into multiple target samples according to the same size and shape;

[0013] Perform three-dimensional cross-section scanning on the multiple target samples by a CT scanner to obtain the soil CT image set;

[0014] Perform filtering processing on the soil CT image set according to a Gaussian filter to reduce the noise level in the soil CT image set and improve the clarity of the soil CT image set;

[0015] Perform contrast processing on the soil CT image set after the filtering processing to adjust the brightness of the soil CT image set and obtain the target CT image set.

[0016] As an optional implementation manner of the first aspect of this application, the method for improving the SegNet network model to obtain an improved SegNet network model is as follows:

[0017] Insert a bottleneck block into both the second encoding module and the third encoding module in the encoder of the SegNet network model;

[0018] Construct a multi-scale feature fusion network, and insert one such multi-scale feature fusion network into each skip connection of the SegNet network model to obtain the improved SegNet network model.

[0019] As an optional implementation manner of the first aspect of this application, the bottleneck block includes two 1×1 depthwise separable convolutional layers and one 3×3 convolutional layer;

[0020] The multi-scale feature fusion network includes a multi-scale feature extraction module, a parallel connection module, a feature enhancement module, a feature revision and fusion module, and a feature aggregation module;

[0021] The multi-scale feature extraction module includes four different convolutional units; the feature revision and fusion module includes three revision units and three extended fusion units.

[0022] As an alternative implementation of the first aspect of the present application, the target CT image set is divided into a training set, a validation set, and a test set according to a preset ratio, and the improved SegNet network model is trained according to the training set and the validation set; specifically:

[0023] The target CT image set is divided into a training set, a validation set, and a test set according to a preset ratio;

[0024] The improved SegNet network model is pre-trained according to the training set;

[0025] The pre-trained improved SegNet network model is verified according to the validation set, and the parameters of the pre-trained improved SegNet network model are tuned according to the verification metrics to complete the training of the improved SegNet network model.

[0026] As an alternative implementation of the first aspect of the present application, the test set is input into the trained improved SegNet network model, and the positions of the plant root pores in the test set are output to complete the detection of the plant root pores in the soil CT image; specifically:

[0027] The test set is input into the trained improved SegNet network model, and the first encoding module of the improved SegNet network model processes the test set to obtain a first encoded feature;

[0028] The second encoding module, the third encoding module, and the fourth encoding module respectively process the output of the previous encoding module and output second encoded features, third encoded features, and fourth encoded features;

[0029] The first encoding module, the second encoding module, the third encoding module, and the fourth encoding module respectively process the corresponding first encoded feature, second encoded feature, third encoded feature, and fourth encoded feature in the corresponding four multi-scale feature fusion networks;

[0030] The four multi-scale feature fusion networks respectively output first multi-scale features, second multi-scale features, third multi-scale features, and fourth multi-scale features to the corresponding first decoding module, second decoding module, third decoding module, and fourth decoding module;

[0031] The fourth decoding module performs feature processing on the fourth multi-scale feature and the fourth encoded feature from the fourth encoding module to obtain a fourth decoded feature;

[0032] The third decoding module, the second decoding module, and the first decoding module respectively perform feature processing on the output of the previous decoding module and the corresponding multi-scale features, and respectively output third decoding features, second decoding features, and first decoding features;

[0033] Output the first decoding feature through the softmax layer in the first decoding module to obtain the position of the plant root pores in the test set, and complete the detection of the plant root pores in the soil CT image.

[0034] As an optional implementation manner of the first aspect of the present application, the processing process of the multi-scale feature fusion network for the first encoded feature; specifically:

[0035] The multi-scale feature extraction module respectively performs feature extraction of different scales on the first encoded feature according to the first convolutional unit, the second convolutional unit, the third convolutional unit, and the fourth convolutional unit, and obtains a first extracted feature, a second extracted feature, a third extracted feature, and a fourth extracted feature;

[0036] Input the second extracted feature, the third extracted feature, and the fourth extracted feature into the parallel connection module for parallel processing to obtain a rough feature;

[0037] Input the rough feature, the first extracted feature, the second extracted feature, the third extracted feature, and the fourth extracted feature into the feature enhancement module for enhancement processing to obtain an enhanced feature, and input the enhanced feature into the three revision units;

[0038] Input the second extracted feature, the third extracted feature, and the fourth extracted feature into the three revision modules respectively. The three revision modules respectively perform feature revision on the second extracted feature, the third extracted feature, and the fourth extracted feature according to the enhanced feature to obtain a second revised feature, a third revised feature, and a fourth revised feature;

[0039] Input the second revised feature, the third revised feature, and the fourth revised feature into the three extended fusion units respectively to obtain a second extended feature, a third extended feature, and a fourth extended feature; fuse the second extended feature, the third extended feature, and the fourth extended feature through one extended fusion unit to obtain an extended revised feature;

[0040] Input the extended revised feature and the enhanced feature into the feature aggregation module for feature aggregation processing to obtain a first multi-scale feature;

[0041] According to the processing process of the multi-scale feature fusion network for the first encoded feature, process the second encoded feature, the third encoded feature, and the fourth encoded feature to obtain a second multi-scale feature, a third multi-scale feature, and a fourth multi-scale feature.

[0042] In a second aspect, an embodiment of the present application provides a detection system for plant root pores in soil CT images based on an improved SegNet model, and the system includes:

[0043] An acquisition module: acquiring a soil CT image set, preprocessing the soil CT image set to obtain a target CT image set;

[0044] An improvement module: improving the SegNet network model to obtain an improved SegNet network model;

[0045] A training module: dividing the target CT image set into a training set, a validation set and a test set according to a preset ratio, and training the improved SegNet network model according to the training set and the validation set;

[0046] A prediction module: inputting the test set into the trained improved SegNet network model, outputting the positions of plant root pores in the test set, and completing the detection of plant root pores in soil CT images.

[0047] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.

[0048] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0049] Compared with the prior art, the present application has the following technical effects:

[0050] (1) By improving the SegNet model, it is possible to more accurately identify plant root pores in soil CT images, reduce misdetection and missed detection situations, and improve detection efficiency and detection accuracy.

[0051] (2) Through reasonable data preprocessing and model training strategies, the improved SegNet model can better adapt to soil CT images from different sources and under different conditions, and improve the generalization ability of the model.

[0052] (3) Using the improved SegNet model for plant root pore detection can replace the traditional manual annotation method, greatly improve the detection efficiency, and reduce the labor cost.

[0053] (4) This method provides a powerful tool for soil structure and plant root system research, helps to reveal the interaction mechanism between plant roots and the soil environment, and promotes the sustainable development of agriculture. Brief Description of the Drawings

[0054] Figure 1 is a flowchart of a method for detecting plant root pores in soil CT images based on an improved SegNet model provided by some embodiments of the present application;

[0055] Figure 2 is a schematic structural diagram of an improved SegNet model in a method for detecting plant root pores in soil CT images based on an improved SegNet model provided by some embodiments of the present application;

[0056] Figure 3 is a schematic structural diagram of a multi-scale feature fusion network in a method for detecting plant root pores in soil CT images based on an improved SegNet model provided by some embodiments of the present application;

[0057] Figure 4 is a schematic structural diagram of a second encoding module in a method for detecting plant root pores in soil CT images based on an improved SegNet model provided by some embodiments of the present application.

[0058] Figure 5 is a detection result diagram of a method for detecting plant root pores in soil CT images based on an improved SegNet model provided by some embodiments of the present application. Detailed Embodiments

[0059] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0060] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects and are not used to describe a specific order or sequence. It should be understood that such data may be interchanged under appropriate circumstances so that the embodiments of the present application can be implemented in an order different from those illustrated or described herein. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the associated objects before and after.

[0061] Next, a method and system for detecting plant root pores in soil CT images based on an improved SegNet model provided by the embodiments of the present application will be described in detail in conjunction with the accompanying drawings through specific embodiments and their application scenarios.

[0062] Embodiment

[0063] A detection method for plant root pores in soil CT images based on an improved SegNet model, comprising:

[0064] S100: Obtain a soil CT image set, preprocess the soil CT image set to obtain a target CT image set;

[0065] It should be noted that S100 specifically is:

[0066] S110: Obtain soil samples from multiple regions, and divide the soil samples into multiple target samples according to the same size and shape;

[0067] S120: Perform three-dimensional cross-section scanning on multiple target samples according to a CT scanner to obtain a soil CT image set;

[0068] S130: Filter the soil CT image set according to a Gaussian filter to reduce the noise level in the soil CT image set and improve the clarity of the soil CT image set;

[0069] S140: Perform contrast processing on the filtered soil CT image set to adjust the brightness of the soil CT image set to obtain a target CT image set.

[0070] Furthermore, collect soil samples from multiple regions. These samples should represent different soil types, structures, water contents, and plant root conditions as much as possible. After collection, divide the soil samples according to a unified size and shape to ensure that each target sample has a consistent size and shape, which is convenient for subsequent scanning and processing;; Standardized soil samples help reduce the inconsistency of scanning results caused by sample differences and provide a reliable basis for subsequent data processing and model training. Use a CT scanner to perform three-dimensional cross-section scanning on each target sample. CT scanning can generate high-resolution images of the internal structure of the soil, including details such as soil particles, pores, and plant roots, providing rich information for subsequent detection of plant root pores. Applying a Gaussian filter to the soil CT image set for filtering processing can smooth the image and reduce noise; Gaussian filtering processing can significantly reduce the noise level in the soil CT image set and improve the clarity of the image, which is crucial for subsequent image analysis and model training because noise may interfere with the recognition ability of the model. Perform contrast adjustment on the filtered soil CT image set. The contrast processing aims to adjust the brightness distribution of the image to make the brightness difference between different regions more obvious; through contrast processing, a clearer and easier-to-identify soil CT image set can be obtained; this not only helps with manual observation and analysis but also provides better input data for subsequent automated detection models.

[0071] S200: Improve the SegNet network model to obtain an improved SegNet network model;

[0072] It should be noted that S200 is specifically as follows:

[0073] S210: Insert a bottleneck block into both the second encoding module and the third encoding module in the encoder of the SegNet network model;

[0074] S220: Construct a multi-scale feature fusion network, and insert a multi-scale feature fusion network into each skip connection of the SegNet network model to obtain an improved SegNet network model.

[0075] Among them, the bottleneck block includes two 1×1 depthwise separable convolutional layers and one 3×3 convolutional layer; the multi-scale feature fusion network includes a multi-scale feature extraction module, a parallel connection module, a feature enhancement module, a feature revision and fusion module, and a feature aggregation module; the multi-scale feature extraction module includes four different convolutional units; the feature revision and fusion module includes three revision units and three extended fusion units.

[0076] Furthermore, in the second encoding module and the third encoding module of the encoder part of the SegNet network model, insert a bottleneck block respectively. The bottleneck block consists of two 1×1 depthwise separable convolutional layers and one 3×3 convolutional layer, aiming to reduce the computational amount, prevent overfitting, and extract deeper features; the bottleneck block reduces the computational complexity of the model by reducing the dimension and number of feature maps. At the same time, the 1×1 convolutional layer helps to fuse features of different channels, while the 3×3 convolutional layer can capture local feature information. This structure enables the model to reduce the computational cost while maintaining high performance. In each skip connection of the SegNet network model, insert a multi-scale feature fusion network. The multi-scale feature fusion network includes a multi-scale feature extraction module, a parallel connection module, a feature enhancement module, a feature revision and fusion module, and a feature aggregation module. Among them, the multi-scale feature extraction module contains four different convolutional units for capturing feature information of different scales; the feature revision and fusion module contains three revision units and three extended fusion units for refining and fusing features; the multi-scale feature fusion network can make full use of feature information of different scales to improve the recognition accuracy of plant root pores by the model. Through the parallel connection and the feature enhancement module, the model can learn more robust feature representations. The feature revision and fusion module further refines and fuses the feature information, making the model more accurate and stable during the detection process. Finally, the feature aggregation module integrates the feature information of different scales and outputs the final detection result.

[0077] S300: Divide the target CT image set into a training set, a validation set, and a test set according to a preset ratio, and train the improved SegNet network model based on the training set and the validation set;

[0078] It should be noted that S300 is specifically as follows:

[0079] S310: Divide the target CT image set into a training set, a validation set, and a test set according to a preset ratio;

[0080] S320: Pre-train the improved SegNet network model based on the training set;

[0081] S330: Validate the pre-trained improved SegNet network model based on the validation set, and perform parameter tuning on the pre-trained improved SegNet network model according to the validation metrics to complete the training of the improved SegNet network model.

[0082] Furthermore, according to a preset ratio (such as 70% training set, 15% validation set, 15% test set), divide the pre-processed target CT image set into three parts: a training set, a validation set, and a test set. The training set is used for the learning process of the model, the validation set is used to monitor the training process and adjust the model parameters, and the test set is used to finally evaluate the model performance; reasonable dataset division can ensure that the model has enough data for learning during the training process, and at the same time avoid overfitting and underfitting; the use of the validation set and the test set helps to evaluate the generalization ability of the model and ensure the performance of the model in practical applications. Use the data in the training set to pre-train the improved SegNet network model; during the pre-training process, the model continuously adjusts the weight parameters through forward propagation and backward propagation algorithms to minimize the difference between the prediction result and the true label; the pre-training stage enables the model to learn the basic feature information in the soil CT image and lay a foundation for subsequent fine-tuning; through continuous iterative training, the model gradually converges and reaches a better initial performance. After pre-training, use the data in the validation set to validate the improved SegNet network model; during the validation process, monitor the performance of the model on the validation set, such as validation metrics such as accuracy, recall rate, F1 score, etc.; according to the validation metrics, tune the parameters of the model, such as adjusting the learning rate, regularization coefficient, batch size, etc., to further improve the performance of the model; the validation and parameter tuning stage can ensure that the model reaches the best performance while maintaining a low overfitting risk. Through continuous parameter adjustment and validation, the model gradually converges to the optimal solution, providing a reliable guarantee for subsequent practical applications.

[0083] S400: Input the test set into the trained improved SegNet network model, output the positions of the plant root pores in the test set, and complete the detection of the plant root pores in the soil CT image;

[0084] It should be noted that S400 specifically is as follows:

[0085] S410: Input the test set into the trained improved SegNet network model. The first encoding module of the improved SegNet network model processes the test set to obtain the first encoded feature;

[0086] S420: The second encoding module, the third encoding module, and the fourth encoding module respectively process the output of the previous encoding module and output the second encoded feature, the third encoded feature, and the fourth encoded feature respectively;

[0087] S430: The first encoding module, the second encoding module, the third encoding module, and the fourth encoding module respectively process the corresponding first encoded feature, second encoded feature, third encoded feature, and fourth encoded feature in the corresponding four multi-scale feature fusion networks;

[0088] S440: The four multi-scale feature fusion networks respectively output the first multi-scale feature, the second multi-scale feature, the third multi-scale feature, and the fourth multi-scale feature to the corresponding first decoding module, second decoding module, third decoding module, and fourth decoding module;

[0089] S450: The fourth decoding module processes the fourth multi-scale feature and the fourth encoded feature from the fourth encoding module to obtain the fourth decoded feature;

[0090] S460: The third decoding module, the second decoding module, and the first decoding module respectively process the output of the previous decoding module and the corresponding multi-scale feature and output the third decoded feature, the second decoded feature, and the first decoded feature respectively;

[0091] S470: Output the first decoded feature through the softmax layer in the first decoding module to obtain the positions of the plant root pores in the test set, and complete the detection of the plant root pores in the soil CT image.

[0092] Furthermore, the soil CT images in the test set are input into the trained improved SegNet network model. First, the first encoding module processes the input test set, extracts the basic features of the image through operations such as convolution, activation, and pooling, and obtains the first encoded feature; the first encoding module can capture the low-level features in the soil CT image, such as edges, textures, etc., providing a basis for subsequent feature extraction and fusion. Then, the second encoding module, the third encoding module, and the fourth encoding module respectively process the output of the previous encoding module. Each encoding module extracts deeper features through similar convolution, activation, and pooling operations, and outputs the second encoded feature, the third encoded feature, and the fourth encoded feature respectively; the hierarchical encoding feature extraction can capture different levels of feature information in the soil CT image, from low-level to high-level, gradually abstracting and refining. The first encoding module, the second encoding module, the third encoding module, and the fourth encoding module respectively input the corresponding encoded features into the corresponding four multi-scale feature fusion networks. The multi-scale feature fusion network extracts and fuses multi-scale features through convolution kernels and feature fusion strategies of different scales, and outputs the first multi-scale feature, the second multi-scale feature, the third multi-scale feature, and the fourth multi-scale feature respectively. The multi-scale feature fusion can make full use of the feature information of different scales, improving the recognition accuracy and robustness of the model for plant root pores. The outputs of the four multi-scale feature fusion networks are respectively transmitted to the corresponding first decoding module, second decoding module, third decoding module, and fourth decoding module. This step provides rich multi-scale feature information for the decoding process, helping the model to more accurately reconstruct and identify plant root pores in the decoding stage. The fourth decoding module first processes the fourth multi-scale feature and the fourth encoded feature from the fourth encoding module, and obtains the fourth decoded feature through operations such as upsampling and convolution. Then, the third decoding module, the second decoding module, and the first decoding module respectively process the output of the previous decoding module and the corresponding multi-scale feature, and output the third decoded feature, the second decoded feature, and the first decoded feature respectively. The hierarchical decoding feature processing can gradually fuse the multi-scale feature information into the decoding process, improving the reconstruction accuracy and recognition ability of the model. Finally, the first decoded feature is processed through the softmax layer in the first decoding module, and the probability map or binary segmentation result of each pixel belonging to the plant root pore is output, so as to obtain the position of the plant root pore in the test set.

[0093] It should be noted that the processing process of the multi-scale feature fusion network for the first encoded feature, the second encoded feature, the third encoded feature, and the fourth encoded feature in S430; specifically:

[0094] S431: The multi-scale feature extraction module performs feature extraction on the first encoded feature at different scales according to the first convolutional unit, the second convolutional unit, the third convolutional unit, and the fourth convolutional unit, obtaining a first extracted feature, a second extracted feature, a third extracted feature, and a fourth extracted feature;

[0095] S432: Input the second extracted feature, the third extracted feature, and the fourth extracted feature into the parallel connection module for parallel processing to obtain a rough feature;

[0096] S433: Input the rough feature, the first extracted feature, the second extracted feature, the third extracted feature, and the fourth extracted feature into the feature enhancement module for enhancement processing to obtain an enhanced feature, and input the enhanced feature into three revision units;

[0097] S434: Input the second extracted feature, the third extracted feature, and the fourth extracted feature into three revision modules respectively. The three revision modules perform feature revision on the second extracted feature, the third extracted feature, and the fourth extracted feature respectively according to the enhanced feature, obtaining a second revised feature, a third revised feature, and a fourth revised feature;

[0098] S435: Input the second revised feature, the third revised feature, and the fourth revised feature into three extended fusion units respectively to obtain a second extended feature, a third extended feature, and a fourth extended feature; fuse the second extended feature, the third extended feature, and the fourth extended feature through an extended fusion unit to obtain an extended revised feature;

[0099] S436: Input the extended revised feature and the enhanced feature into the feature aggregation module for feature aggregation processing to obtain a first multi-scale feature;

[0100] S437: According to the steps of processing the first encoded feature in S431 - S436, process the second encoded feature, the third encoded feature, and the fourth encoded feature to obtain a second multi-scale feature, a third multi-scale feature, and a fourth multi-scale feature.

[0101] Further, the multi-scale feature extraction module uses the first convolutional unit, the second convolutional unit, the third convolutional unit, and the fourth convolutional unit to perform feature extraction on the first encoded feature at different scales respectively, obtaining the first extracted feature, the second extracted feature, the third extracted feature, and the fourth extracted feature, which respectively represent information at different scales. The second extracted feature, the third extracted feature, and the fourth extracted feature are input into the parallel connection module for parallel processing to obtain rough features. These features are obtained by parallel processing the extracted features at different scales and contain richer information. The rough features and all the extracted features (the first extracted feature, the second extracted feature, the third extracted feature, and the fourth extracted feature) are input into the feature enhancement module for enhancement processing to obtain enhanced features, which will be used for feature revision and extended fusion in subsequent steps. At the same time, the enhanced features are also input into three revision units for subsequent use. The three revision modules perform feature revision on the second extracted feature, the third extracted feature, and the fourth extracted feature respectively according to the enhanced features, obtaining the second revised feature, the third revised feature, and the fourth revised feature. These revised features are more accurate and stable. The second revised feature, the third revised feature, and the fourth revised feature are respectively input into three extended fusion units, and one extended fusion unit fuses these three extended features. The extended revised feature is obtained, which is obtained by fusing the revised features at different scales and contains broader and deeper information. The extended revised feature and the enhanced features are input into the feature aggregation module for feature aggregation processing to obtain the first multi-scale feature, which is obtained by aggregating the features at different scales and different processing stages and has stronger representation ability. According to the steps of processing the first encoded feature in S431 - S436, the same processing is performed on the second encoded feature, the third encoded feature, and the fourth encoded feature, obtaining the second multi-scale feature, the third multi-scale feature, and the fourth multi-scale feature, which respectively represent the information fusion results of different encoded features at different scales.

[0102] A method for detecting plant root pores in soil CT images based on an improved SegNet model according to this embodiment. By obtaining soil samples from multiple regions and performing standardized segmentation, a high-quality soil CT image set is obtained using a CT scanner. The Gaussian filter is used to filter the images, effectively reducing the noise level and improving the image clarity. By adjusting the image brightness through contrast processing, the quality of the target CT image set is further optimized, providing a reliable basis for subsequent detection. A bottleneck block is inserted into the encoder of the SegNet network model to enhance the model's expression ability and feature extraction ability. A multi-scale feature fusion network is constructed and inserted to achieve effective extraction, enhancement, revision, and fusion of features at different scales, improving the model's understanding and recognition ability of complex image structures. The training set, validation set, and test set are reasonably divided according to a preset ratio to ensure the comprehensiveness and effectiveness of model training. Through pre-training and validation set verification, the model parameters are optimized in combination with the validation metrics to ensure the stability and accuracy of the model. In the test stage, the improved SegNet network model processes the test set through the multi-scale feature fusion network to accurately output the positions of plant root pores. The entire detection process is efficient and automated, reducing manual intervention and improving the detection efficiency and accuracy.

[0103] In March 2024, the applicant applied the detection method to the soil system under the condition of crop planting in the lime concretion black soil in the North China Plain, and successfully detected the spatial position of root pores in the soil (see Figure 5 shown), and the cutting accuracy reached more than 96%, which is sufficient to support the application of most scientific research.

[0104] It should be noted that for a method for detecting plant root pores in soil CT images based on an improved SegNet model provided in the embodiment of the present application, the execution subject can be a detection system for plant root pores in soil CT images based on an improved SegNet model, or a control module in the detection system for plant root pores in soil CT images based on an improved SegNet model for executing the method for detecting plant root pores in soil CT images based on an improved SegNet model. In the embodiment of the present application, taking a detection system for plant root pores in soil CT images based on an improved SegNet model to execute the method for detecting plant root pores in soil CT images based on an improved SegNet model as an example, the method for detecting plant root pores in soil CT images based on an improved SegNet model provided in the embodiment of the present application is described.

[0105] A detection system for plant root pores in soil CT images based on an improved SegNet model, comprising:

[0106] Acquisition module: Acquire a set of soil CT images, preprocess the set of soil CT images to obtain a set of target CT images;

[0107] Improvement module: Improve the SegNet network model to obtain an improved SegNet network model;

[0108] Training module: Divide the set of target CT images into a training set, a validation set, and a test set according to a preset ratio, and train the improved SegNet network model according to the training set and the validation set;

[0109] Prediction module: Input the test set into the trained improved SegNet network model, output the positions of the plant root pores in the test set, and complete the detection of the plant root pores in the soil CT images.

[0110] A detection system for plant root pores in soil CT images based on an improved SegNet model in an embodiment of the present application may be a device, or a component, an integrated circuit, or a chip in a terminal. The device may be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a palmtop computer, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic device may be a server, a Network Attached Storage (NAS), a personal computer (PC), etc. The embodiments of the present application do not make specific limitations.

[0111] A detection method system for plant root pores in soil CT images based on an improved SegNet model in an embodiment of the present application may be a device with an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems. The embodiments of the present application do not make specific limitations.

[0112] A detection system for plant root pores in soil CT images based on an improved SegNet model provided by an embodiment of the present application can implement Figures 1 to 4 each process implemented by a detection method for plant root pores in soil CT images based on an improved SegNet model in the method embodiment. To avoid repetition, it will not be elaborated here.

[0113] A detection system for plant root pores in soil CT images based on an improved SegNet model according to this embodiment. The acquisition module can collect and process a large number of soil CT image sets, and improve the image quality through preprocessing steps (such as denoising, enhancing contrast, etc.) to obtain a target CT image set suitable for subsequent detection, which helps to improve the accuracy and reliability of detection. The improvement module makes customized improvements to the SegNet network model, enabling the improved SegNet network model to more effectively capture the feature information in soil CT images, especially the subtle features of plant root pores, which helps to improve the detection accuracy and generalization ability of the model. The training module can divide the target CT image set into a training set, a validation set, and a test set according to a preset ratio to ensure that the training process of the model is both comprehensive and efficient; by training the improved SegNet network model with the training set and the validation set and using the validation set for model tuning, it can ensure that the model achieves the best performance on the test set. This systematic training process helps to improve the stability and accuracy of the model. The prediction module can input the test set into the trained improved SegNet network model and quickly and accurately output the positions of plant root pores in the test set. This benefits from the powerful feature extraction and image segmentation capabilities of the improved SegNet network model, enabling the system to accurately identify plant root pores in complex soil CT images.

[0114] Optionally, an embodiment of the present application further provides an electronic device, including a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, it implements each process of the above-mentioned embodiment of the detection method for plant root pores in soil CT images based on an improved SegNet model and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0115] An embodiment of the present application further provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by the processor, it implements each process of the above-mentioned embodiment of the detection method for plant root pores in soil CT images based on an improved SegNet model and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0116] Among them, the processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc.

[0117] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in a reverse order according to the functions involved. For example, the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, the features described with reference to certain examples may be combined in other examples.

[0118] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described example methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present application.

[0119] The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present application, those of ordinary skill in the art can also make many forms without departing from the purpose of the present application and the scope protected by the claims, and all of them fall within the protection scope of the present application.

Claims

1. A method for detecting plant root pores in soil CT images based on an improved SegNet model, characterized in that: The method comprises: Acquire a soil CT image set, and preprocess the soil CT image set to obtain a target CT image set; The SegNet network model is improved to obtain an improved SegNet network model; Dividing the target CT image set into a training set, a validation set, and a test set according to a preset ratio, and training the improved SegNet network model according to the training set and the validation set; The test set is input into the trained improved SegNet network model, and the positions of the plant root pores in the test set are output to complete the detection of the plant root pores in the soil CT image.

2. According to claim 1, a method for detecting plant root pores in soil CT images based on an improved SegNet model is characterized in that: The step of acquiring a soil CT image set and preprocessing the soil CT image set to obtain a target CT image set is as follows: Acquire soil samples from multiple regions, and divide the soil samples into multiple target samples of the same size and shape; Performing three-dimensional cross-sectional scanning on the plurality of target samples using a CT scanner to obtain the soil CT image set; Performing filtering processing on the soil CT image set according to a Gaussian filter to reduce the noise level in the soil CT image set and improve the clarity of the soil CT image set; Contrast processing is performed on the soil CT image set after the filtering process to adjust the brightness of the soil CT image set to obtain the target CT image set.

3. According to claim 2, a method for detecting plant root pores in soil CT images based on an improved SegNet model is characterized in that: The SegNet network model is improved to obtain an improved SegNet network model; specifically: Inserting a bottleneck block into the second encoding module and the third encoding module in the encoder of the SegNet network model; A multi-scale feature fusion network is constructed, and one of the multi-scale feature fusion networks is inserted into each jump connection of the SegNet network model to obtain the improved SegNet network model.

4. The method for detecting plant root pores in soil CT images based on the improved SegNet model according to claim 3, characterized in that: The bottleneck block includes two 1×1 depthwise separable convolutional layers and one 3×3 convolutional layer; The multi-scale feature fusion network includes a multi-scale feature extraction module, a parallel connection module, a feature enhancement module, a feature revision fusion module and a feature aggregation module; The multi-scale feature extraction module includes four different convolution units; the feature revision fusion module includes three revision units and three extension fusion units.

5. The method for detecting plant root holes in soil CT images based on the improved SegNet model according to claim 1, characterized in that: The target CT image set is divided into a training set, a validation set and a test set according to a preset ratio, and the improved SegNet network model is trained according to the training set and the validation set; Specifically: Dividing the target CT image set into a training set, a validation set, and a test set according to a preset ratio; Pre-training the improved SegNet network model according to the training set; The pre-trained improved SegNet network model is verified according to the verification set, so as to perform parameter tuning on the pre-trained improved SegNet network model according to the verification index, and complete the training of the improved SegNet network model.

6. The method for detecting plant root holes in soil CT images based on the improved SegNet model according to claim 4, characterized in that: The test set is input into the trained improved SegNet network model, the positions of the plant root pores in the test set are output, and the detection of the plant root pores in the soil CT image is completed; specifically: The test set is input into the trained improved SegNet network model, and the first encoding module of the improved SegNet network model processes the test set to obtain a first encoding feature; The second encoding module, the third encoding module and the fourth encoding module process the output of the previous encoding module respectively, and output the second encoding feature, the third encoding feature and the fourth encoding feature respectively; The first encoding module, the second encoding module, the third encoding module and the fourth encoding module respectively process the corresponding first encoding feature, the second encoding feature, the third encoding feature and the fourth encoding feature in the corresponding four multi-scale feature fusion networks; The four multi-scale feature fusion networks output the first multi-scale feature, the second multi-scale feature, the third multi-scale feature and the fourth multi-scale feature to the corresponding first decoding module, the second decoding module, the third decoding module and the fourth decoding module respectively; The fourth decoding module performs feature processing on the fourth multi-scale feature and the fourth encoding feature from the fourth encoding module to obtain a fourth decoding feature; The third decoding module, the second decoding module and the first decoding module respectively perform feature processing on the output of the previous decoding module and the corresponding multi-scale features, and output a third decoding feature, a second decoding feature and a first decoding feature respectively; The first decoding feature is output through the softmax layer in the first decoding module to obtain the position of the plant root pores in the test set, thereby completing the detection of the plant root pores in the soil CT image.

7. The method for detecting plant root holes in soil CT images based on the improved SegNet model according to claim 6, characterized in that: The multi-scale feature fusion network processes the first coding feature; specifically: The multi-scale feature extraction module extracts features of different scales on the first coding feature according to the first convolution unit, the second convolution unit, the third convolution unit and the fourth convolution unit, respectively, to obtain a first extracted feature, a second extracted feature, a third extracted feature and a fourth extracted feature; Inputting the second extracted features, the third extracted features and the fourth extracted features into the parallel connection module for parallel processing to obtain rough features; Input the rough feature and the first extracted feature, the second extracted feature, the third extracted feature and the fourth extracted feature into the feature enhancement module for enhancement processing to obtain enhanced features, and input the enhanced features into the three revision units; The second extracted feature, the third extracted feature, and the fourth extracted feature are respectively input into the three revision modules, and the three revision modules respectively revise the second extracted feature, the third extracted feature, and the fourth extracted feature according to the enhanced feature to obtain the second revised feature, the third revised feature, and the fourth revised feature; Inputting the second revision feature, the third revision feature and the fourth revision feature into the three extension fusion units respectively to obtain the second extension feature, the third extension feature and the fourth extension feature; fusing the second extension feature, the third extension feature and the fourth extension feature through one extension fusion unit to obtain the extended revision feature; Inputting the extended revision feature and the enhanced feature into the feature aggregation module for feature aggregation processing to obtain a first multi-scale feature; According to the processing process of the first coding feature by the multi-scale feature fusion network, the second coding feature, the third coding feature and the fourth coding feature are processed to obtain a second multi-scale feature, a third multi-scale feature and a fourth multi-scale feature.

8. A detection system for plant root pores in soil CT images based on an improved SegNet model, capable of implementing a detection method for plant root pores in soil CT images based on an improved SegNet model as described in any one of claims 1 to 7, characterized in that: The system comprises: Acquisition module: acquiring a soil CT image set, preprocessing the soil CT image set, and obtaining a target CT image set; Improvement module: improve the SegNet network model to obtain an improved SegNet network model; Training module: dividing the target CT image set into a training set, a validation set and a test set according to a preset ratio, and training the improved SegNet network model according to the training set and the validation set; Prediction module: input the test set into the trained improved SegNet network model, output the position of the plant root pores in the test set, and complete the detection of the plant root pores in the soil CT image.

9. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of a method for detecting plant root pores in soil CT images based on an improved SegNet model as described in any one of claims 1 to 7 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of a method for detecting plant root pores in soil CT images based on an improved SegNet model as described in any one of claims 1 to 7 are implemented.

Images