A method of protecting microtubules from rodent destruction

Abstract

The present application belongs to the technical field of plant root monitoring, and particularly relates to a method for protecting micro-root tubes from being damaged by rodents, which comprises the following steps: identifying specific positions of the micro-root tubes through micro-root tube data acquisition, image classification training, contour feature extraction, alarm identification and video pre-recording, and taking snapshots; using a key algorithm of frame separation to analyze, so as to realize protection of the micro-root tubes and automatic identification and automatic alarm of the micro-root tubes damaged or lost by rodents; after the micro-root tubes are determined and identified, an aluminum alloy pop-top can with a removed top end is sleeved on an exposed end; since the material of the pop-top can is plastic, after the root system is determined, the pop-top can can be pinched to make the pop-top can more closely attached to the micro-root tube, so as to play a role in protecting the micro-root tube.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of plant root monitoring, and particularly relates to a method for protecting micro-root tubes from being damaged by rodents. BACKGROUND

[0002] Plant roots are the dynamic interface between soil and plants, and it is relatively difficult to study them in the process of growing in the soil. Roots are usually taken out destructively for statistical analysis, so it is difficult to continuously monitor them. The birth and application of the micro-root tube technology further develop the means for studying plant roots. The micro-root tube is a device for fixing in soil to monitor root growth. When the micro-root tube is set, 25 cm is left on the ground, and there is a rubber soft cover at the micro-root tube port to protect the micro-root tube from entering water. However, in winter, snow covers the part of the micro-root tube exposed on the ground, and rodents (mice) gnaw the rubber soft cover, causing the micro-root tube to enter water, and finally the micro-root tube is abandoned.

[0003] In order to protect the rubber soft cover of the micro-root tube from being gnawed by mice, so as to better realize long-term monitoring of the growth and development trend of a single root system or a certain root fragment, and to conduct detailed research on the characteristics of plant root growth, death and decomposition, the application provides a method for protecting micro-root tubes from being damaged by rodents. SUMMARY

[0004] The application provides a method for protecting micro-root tubes from being damaged by rodents. After the micro-root tube roots are measured and identified, an aluminum alloy pop-top can without a top end is sleeved on the exposed end of the rubber soft cover of the micro-root tube and pinched, so that the technical problem of the micro-root tube port being easily gnawed by rodents is solved.

[0005] In order to achieve the above purpose, the technical scheme adopted by the application is as follows: a method for protecting micro-root tubes from being damaged by rodents, specifically comprising the following steps:

[0006] 1) Data acquisition: a large amount of image data of the micro-root tubes in the fixed sample plot with video monitoring is collected to form an image data comparison library;

[0007] 2) Classification training: the micro-root tube images in the database are identified, and an AdBoost classifier algorithm is used for classifier training;

[0008] 3) Contour feature extraction: first, the micro-root tube is separated from the background, the target contour line is extracted, and curve fitting is performed;

[0009] 4) Alarm identification: the image is identified for abnormal result information, and the interface is sent to the front-end interface to display the alarm information and give a voice reminder;

[0010] 5) Video pre-recording: save a segment of video data before the alarm and a segment of video data after the alarm for subsequent viewing and analysis of the cause;

[0011] 6) Set up a pop can: a pop can made of aluminum alloy with the top end removed is set on the exposed end of the rubber hose of the micro root canal and pinched.

[0012] The exposed ground end of the micro root canal is 25 cm.

[0013] Further, in step 1), each image in the image data comparison library is a target effect sample.

[0014] Further, in step 1), the feature extraction of the image adopts the HOG algorithm.

[0015] Further, in step 2), the image is captured in real time by a camera, the frequency of the snapshot is 30 Hz / s, 3-5 Hz / s is taken per second, the time is 10 s, and about 40 Hz is taken per group.

[0016] Further, in step 3), the curve fitting method is the least square method.

[0017] Further, in step 4), the abnormal result is the deviation of 5-6 Hz data continuously or at intervals, and the result is obtained every 10 s of data.

[0018] Further, in step 1), the image is divided into 8x8 small grids, and the gradient direction histogram of the image in each small grid is calculated.

[0019] Further, in step 4), the image recognition adopts the faster R-CNN algorithm.

[0020] Further, the pop can is made of aluminum alloy and has a length of 15 cm.

[0021] Compared with the prior art, the present application has the following beneficial effects:

[0022] 1. The present application identifies the specific position of the micro root canal through image recognition, captures the micro root canal, uses the key algorithm analysis of frame separation flow, realizes the protection of the micro root canal, and automatically identifies and automatically alarms the micro root canal damaged or lost by rodents.

[0023] 2. The present application adopts infrared thermal imaging technology, converts the thermal signal scanned to the micro root canal into a stereoscopic image, and realizes real-time monitoring of the position of the micro root canal; in a specific position, it is sampled and photographed, and an intelligent analysis algorithm is run in the background, when the rodent occurs, it timely monitors and sends an alarm information. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1Flow chart of the method of the present application;

[0025] Figure 2 Flow chart of the data collection of the present application. DETAILED DESCRIPTION

[0026] The present application will be described in detail below with specific examples. The following examples will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present application. These are within the scope of protection of the present application.

[0027] Example 1

[0028] The present application proposes a method for protecting micro root tubes from being damaged by rodents. After identifying and measuring the root system of the micro root tube in the fixed sample plot with video monitoring, a top-removable aluminum alloy pop can (e.g., a beer pop can) is fitted on the exposed end, with a length of 15 cm. Since the material of the pop can is easy to deform, after the root system is measured, the pop can can be pinched to make the pop can fit the micro root tube better, thereby protecting the micro root tube.

[0029] Specifically, the following steps are included:

[0030] The image meeting the recognition algorithm is more effectively and quickly extracted by the frame separation flow method, and then the image is subjected to feature extraction and recognition, and then a certain image processing technique is used to reduce the detection range of the target, further improving the accuracy and performance of the algorithm and reducing the error of the algorithm. The final effect is a snapshot effect map - a contour feature map - a binary value map.

[0031] 1. Data collection

[0032] First, the image is snapped, a large amount of image data is collected, and an image data comparison library is formed. The more comprehensive the collected image data, the higher the accuracy of the system recognition. Each snapped image is a target effect sample. After the target effect feature is extracted, a target contour map is generated, and the target contour map is subjected to image processing to extract a binary value map of the target object. The information data of all sample image processing constitutes a complete database for image recognition.

[0033] In the data collection part, the HOG algorithm is used for feature extraction of the image:

[0034] The algorithm flow is as follows:

[0035] HOG (Histogram of Oriented Gradient) feature is a common feature extraction algorithm in object recognition and pattern matching, which is based on local pixel block to extract feature histogram. The object local deformation and light influence have good stability. HOG feature is initially used to identify portrait. Through HOG feature extraction + SVM training, good results can be obtained. The approximate process of HOG feature extraction is as follows:

[0036] First step:

[0037] Gamma correction is mainly to correct the input image, mainly to compensate for the gray deviation caused by the display. The common coefficient is about 2.5.

[0038] Second step:

[0039] Image to grayscale, which belongs to the conventional technical means.

[0040] Third step:

[0041] Calculate the gradient and direction of the image, use SOBEL operator to realize, finally get the gradient amplitude and angle of the image.

[0042] Gradient calculation-formula principle:

[0043] Vertical direction Horizontal direction

[0044] Gradient

[0045] Fourth step:

[0046] Divide the image into 8x8 small grids, and make gradient direction histogram for the image in each small grid. Each 8x8 = 64 pixels is a CELL. According to the angle, each CELL is divided into 9 histogram blocks (BIN). The range of each BIN is 20 degrees. Assuming that the angle of a pixel point in the CELL is 10, the corresponding gradient value is accumulated and put into the first histogram numbered 0. Finally, the data of each histogram numbered 0-8 is obtained. Then according to the angle value, the weight of the gradient of the corresponding pixel on the adjacent histogram is calculated, and the corresponding value is accumulated to the adjacent histogram according to the weight. In this way, the most important step of HOG, weight angle histogram data statistics, is completed.

[0047] 2. Classification training

[0048] For the micro root canal image to be identified, the Adboost classifier algorithm is used for classifier training, and the generated classifier file is saved in an XML file. The database template is established and put into operation in the monitoring system. During system operation, the SDK camera performs real-time video acquisition and real-time video display, and decodes each frame of data. After the real-time data is transmitted to the algorithm analysis library, the algorithm first performs preprocessing on the current image data and compares and analyzes the previously trained database. When image data that meets the recognition algorithm is detected, the picture is captured. The capture frequency is 30 frames / s, 3-5 frames / s are taken every second, the time is 10 s, about 40 frames of pictures are taken every group, and each frame of picture is subjected to subsequent segmentation, feature extraction and other algorithm operations.

[0049] The classification training uses the Adboost classifier algorithm:

[0050] The algorithm flow is as follows:

[0051] Given the training data set: (x1, y1), … (x N , y N ), where y i ∈{-1, 1} is used to represent the class label of the training sample, i = 1, …, N. The purpose of Adboost is to learn a series of weak classifiers or basic classifiers from the training data, and then combine these weak classifiers into a strong classifier.

[0052] Related symbol definitions:

[0053] Di(i): weight distribution of the training sample set;

[0054] W i : weight of each training sample;

[0055] h: weak classifier;

[0056] H: basic classifier;

[0057] H final : final strong classifier;

[0058] e: error rate;

[0059] α t : weight of weak classifier;

[0060] The algorithm flow of Adboost is as follows:

[0061] (1) First, initialize the weight distribution of the training data, and initially assign the same weight to each training sample, w i = 1 / N, so that the initial weight distribution Di(i) of the training sample set.

[0062]

[0063] (2) Iteration t = 1,..., T

[0064] (a) Select a weak classifier h with the lowest current error rate as the t-th base classifier Ht, and calculate the error of the weak classifier ht; X→{-1, 1} on the distribution Di:

[0065]

[0066] (b) Calculate the proportion of the weak classifier in the final classifier (the weight of the weak classifier is denoted by a):

[0067]

[0068] (c) Update the weight distribution Di+1 of the training samples:

[0069] where

[0070] (3) Finally, according to the weak classifier weight a t Combine each weak classifier, i.e.

[0071]

[0072] Through the sign function, a strong classifier is obtained:

[0073]

[0074] Related explanations:

[0075] Because the weight update depends on a t And a depends on the error rate e, so we can directly express the weight update formula with e. Sample weight update formula:

[0076] where

[0077] (1) When the sample is misclassified, y i H t (x i ) = -1

[0078]

[0079] (2) When the sample is correctly classified, y i H t (x i ) = 1

[0080]

[0081] Based on the above derivation, the formula for updating the weights of the sample classification and time is:

[0082] Error classification sample, weight update:

[0083] Correct classification sample, weight update:

[0084] 3. Feature extraction

[0085] Contour feature extraction, first segmentation, the shovel and the background is divided, secondly, the target contour line extraction, so as to get the target contour line graph. After obtaining the target contour line graph, the least square method is used for curve fitting. After fitting, the curve condition of each micro root canal is obtained, and whether the micro root canal is damaged by rodents is judged according to the length and height of the curve and the system setting threshold. Continuous or interval 5-6 data deviation can be considered as abnormal condition, and each group of data is within 10s.

[0086] Image recognition uses faster R-CNN algorithm:

[0087] The algorithm process is as follows:

[0088] (1) First, the image is scaled to a fixed size M*N, and then the image is sent into the network.

[0089] (2) Feature map (Feature Map) is extracted by VGG network, and effective anchors are found.

[0090] (3) Feature map (Feature Map) is obtained by RPN network (Region Proposal Networks) effective Anchor confidence and coordinate system coefficient with prediction box. 2K (positive sample and negative sample x K anchor coefficient) scores and 4K (anchor coordinate offset coefficient x K anchor coefficient) coordinates.

[0091] (4) The feature map and the prediction box calculated by the Proposal are obtained by ROIPooling to get the fixed size prediction target feature map, that is, the target is taken out from the feature map by using the prediction box.

[0092] 4. Alarm identification

[0093] The abnormal result information is sent to a front-end interface or other systems through an interface. After receiving the alarm information, the alarm module displays the alarm information and gives voice reminders, and the specific alarm mode can be in multiple modes.

[0094] 5. Video pre-recording

[0095] The alarm module triggers the snapshot storage module to perform data snapshot and video recording functions. The video pre-recording function is adopted, which can save a certain period of video data before and after the alarm, thereby facilitating subsequent viewing and analyzing the causes.

[0096] 6. Easy-open can

[0097] The micro root tube has an outer diameter of 5.0 cm. In order to protect the rubber soft cover of the micro root tube from being gnawed by mice, an aluminum alloy easy-open can (for example, a beer easy-open can) with the top end removed is sleeved on the exposed end, and the length is 15 cm. Since the material of the easy-open can is plastic, after the root system is measured, the easy-open can can be pinched to make the easy-open can and the micro root tube fit better, thereby better protecting the micro root tube.

[0098] The specific embodiments of the present application are described above. It should be understood that the present application is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the present application. In the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method of protecting a micro-root canal from destruction by rodents, comprising: Specifically comprising the following steps: 1) micro root canal data acquisition: a large amount of image data acquisition is carried out on the micro root canal arranged in the video monitoring fixed sample plot to form an image data comparison library, and a target contour map is generated after target effect map features are extracted; 2) image classification training: the micro root canal images in the database are identified, and an AdBoost classifier algorithm is used for classifier training; 3) contour feature extraction: first, segmentation is carried out, the micro root canal is separated from the background, the target contour line is extracted and curve fitting is carried out; 4) alarm identification: abnormal result information of the image is identified, and is sent to the front-end interface through an interface to display alarm information and voice prompt; 5) video pre-recording: a section of video data before the alarm and a section of video data after the alarm are saved for subsequent viewing and analysis; 6) easy open can setting: the micro root canal which has not alarmed is sleeved with an easy open can with the top end removed and is pinched.

2. A method of protecting a micro-root canal from destruction by rodents according to claim 1, wherein: The micro root canal exposed ground end is 25 cm.

3. The method of protecting a micro-root canal from destruction by a rodent according to claim 1, wherein: In step 1), each image in the image data comparison library is a target effect map sample.

4. A method of protecting a micro-root canal from destruction by a rodent in accordance with claim 3, wherein: In step 1), the HOG algorithm is used for feature extraction of the image.

5. The method of protecting a micro-root canal from destruction by a rodent according to claim 1, wherein: In step 2), the image is captured by a camera in real time, the snapshot frequency is 30 frames / s, 3-5 frames / s are taken per second, the time is 10 s, and 40 frames are taken per group.

6. The method of protecting a micro-root canal from destruction by a rodent according to claim 1, wherein: In step 3), the curve fitting method is the least square method.

7. The method of protecting a micro-root canal from destruction by a rodent according to claim 1, wherein: In step 4), the abnormal result is a deviation of 5-6 frames of continuous or interval data, and the result is obtained every 10 s of data.

8. The method of protecting a micro-root canal from destruction by a rodent according to claim 4, wherein: In step 1), the image is divided into 8x8 small grids, and a gradient direction histogram is made for the image in each small grid.

9. A method of protecting a micro-root canal from destruction by rodents according to claim 7, wherein: In step 4), the image recognition uses the faster R-CNN algorithm.

10. The method of protecting a micro-root canal from destruction by a rodent according to claim 1, wherein: The easy open can is made of aluminum alloy material, and the length is 15 cm.

Images