Efficient sugarcane stalk node recognition method based on variational modal decomposition

Abstract

The present application relates to sugarcane planting technical field, especially based on the efficient sugarcane node recognition method of variational mode decomposition, including the following steps: the original photoelectric signal of the vertical projection image of the obtained sugarcane is obtained through linear array CCD sensor;The original photoelectric signal is carried out fixed threshold binaryzation, to obtain the outline signal of sugarcane;The outline signal is decomposed into several modal components, and each modal component is carried out HHT transformation, to obtain the HHT marginal spectrum corresponding to each modal component;Select feature signal, and the saturation value of the feature signal after normalization sets node threshold value;The wave crest greater than the node threshold value in the feature signal after normalization is obtained, to obtain node wave crest, and the position corresponding to the node wave crest is used as the node position of sugarcane.The present application can reduce the cost input of hardware and reduce the difficulty of subsequent data processing;It does not need to produce physical contact to sugarcane, reduces the damage to sugarcane bud.

Description

Technical Field

[0001] This invention relates to the field of sugarcane planting technology, and in particular to an efficient sugarcane stem node identification method based on variational mode decomposition. Background Technology

[0002] Sugarcane is the most basic raw material for the global sugar industry, accounting for over 92% of China's total sugar production. However, due to the low level of mechanization in China's sugarcane production, its sugar prices are not competitive compared to other countries, such as the United States, Brazil, India, and Thailand. Therefore, the mechanization of sugarcane cultivation plays a crucial role in improving the competitiveness of China's sugar industry. Pre-cutting sugarcane seeds is widely used in major hilly sugarcane-producing areas due to the small size of the planter, high seed uniformity, and high germination rate. Since sugarcane buds all grow at the nodes, pre-cutting is necessary during planting. This involves cutting the sugarcane into segments with a certain number of buds. The cuts should avoid the nodes as much as possible to prevent damage to the buds and maintain the germination rate. Automated production of pre-cut sugarcane seeds is a key factor in effectively improving the efficiency of this planting method. In order to reduce the damage to the seed buds caused by the cutting knife during the subsequent sugarcane cutting process, it is necessary to quickly and accurately identify and locate the sugarcane stem nodes, thereby improving the germination rate of pre-cut sugarcane seeds and the sugarcane yield per unit area of ​​land.

[0003] Sugarcane feature recognition can be divided into two types: contact and non-contact, depending on the data acquisition method. While contact detection can disregard the sugarcane's color and surface wax powder, many researchers prefer non-contact detection because it avoids damage to the sugarcane buds. It utilizes optical principles, acquiring information through industrial cameras and photoelectric sensors.

[0004] Traditionally, many researchers convert captured RGB images into HSV images for further processing, as HSV images are more suitable for human perception. Shangping Lu et al. (2010) extracted the S and H components to generate a synthetic image and used the obtained feature indices to build an SVM model for cluster analysis; Jiqing Chen et al. (2021) extracted the S component for vertical decomposition and located nodes based on the minimum sum of local pixels. Pothula et al. (2014) proposed using the unit area under a single normalized gray value curve as the best indicator for identifying nodes and inter-nodes. However, most of these methods require a single or specific background to reduce the difficulty of image processing. Xiao Hu et al. (2019) obtained edge probability images by using a structured random forest method and applied the heuristic algorithm to multiple binarized images. Although this method has been used to deal with various backgrounds, its effectiveness in complex production environments is still insufficient. Rui Yang et al. (2020) used features of average gradient and variance gradient, but it requires two horizontally opposed cameras to obtain the entire original image. Weizheng Zhang et al. (2017) were the first to apply hyperspectral imaging technology to solve the problems of similar colors between sugarcane nodes and stems, as well as interference from white fruit bloom. The above methods are for static recognition, and some researchers have conducted more research on dynamic recognition. Brajesh Nare et al. (2019) used the Sobel operator to extract edge information from binarized images and conducted related research on the throughput of sugarcane seeds. Zhou et al. (2020) used the Sobel operator to calculate the horizontal R component and extract obvious gradient feature vectors at leaf scars. However, traditional image processing methods are almost incompatible with production environments and the requirements of dynamic recognition. Therefore, with the improvement of computing power, deep learning-based object detection methods are increasingly being applied to the field of sugarcane node recognition. Moshashai et al. (2008) extracted edge contour information through convolution operations based on the right-side Sobel edge correction mask matrix. Shangping Li et al. (2020) constructed a lightweight YOLO v3 network, reducing the response time to 28ms and improving the efficiency of real-time dynamic sugarcane node identification. They also innovatively studied the impact of external factors on the algorithm's recognition accuracy. Wen Chen et al. (2021) used five different deep learning frameworks combined with data augmentation methods to identify and analyze complex real-world planting environments at different stages, and experimentally verified the superiority of YOLO v4. However, to achieve higher real-time performance and recognition efficiency, these methods require greater hardware investment, which is not conducive to reducing sugar production costs.

[0005] To address the bottlenecks encountered in non-contact detection, Meng et al. (2019) proposed using photoelectric sensors combined with signal processing technology: two laser rangefinders were used to detect the sugarcane contour signal on the surface, and the signal was decomposed into 8 layers using the db5 wavelet function. Then, different thresholds were applied to signals where the node distribution could be clearly observed, and a multi-sensor redundancy algorithm based on Gaussian membership function was combined with a signal preservation algorithm to synthesize the sugarcane nodes and specific locations. The low-dimensional signals acquired by one or more photoelectric sensors can be processed at high speed by a PC with average performance and the results can be returned. This reduces the investment in hardware costs and the workload of data acquisition and processing, and does not cause additional damage to the sugarcane, providing a new approach for efficient and low-cost non-destructive detection of sugarcane nodes. The above methods can identify nodes to a certain extent, but there are still some problems: (1) Most of them are based on static research, and the imaging range of a single acquisition device is limited. To meet the conditions of continuous and dynamic efficient real-time detection, more costs need to be invested; (2) There is little research on the impact of external factors on the algorithm's recognition performance; (3) The recognition time is long and the efficiency is low. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides an efficient sugarcane stem node identification method based on variational mode decomposition, which can reduce hardware costs and simplify subsequent data processing; it also eliminates the need for physical contact with the sugarcane, thus reducing the risk of damage to sugarcane buds.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] An efficient sugarcane stem node identification method based on variational mode decomposition includes the following steps:

[0009] S1. A vertical projection image of the sugarcane is obtained by scanning with a parallel light source parallel to the radial direction of the sugarcane, and the vertical projection image is used to obtain the original photoelectric signal through a linear CCD sensor.

[0010] S2. Binarize the original photoelectric signal from step S1 with a fixed threshold to obtain the outline signal of the sugarcane.

[0011] S3. The contour signal obtained in step S2 is decomposed into several modal components by the VMD algorithm, and HHT transformation is performed on each modal component to obtain the HHT marginal spectrum corresponding to each modal component.

[0012] S4. Select the HHT marginal spectrum of the maximum energy range of the low-frequency part of the signal in step S3 as the characteristic signal of the sugarcane node, and normalize the characteristic signal to set the node threshold according to the saturation value of the normalized characteristic signal.

[0013] S5. Obtain the peaks in the normalized feature signal that are greater than the node threshold in step S4 to obtain the node peaks, and take the positions corresponding to the node peaks as the sugarcane node positions.

[0014] Further, in step S2, element thresholds are obtained through Otsu threshold selection, and the original photoelectric signal is preprocessed according to binarization to obtain the occluded elements in each element threshold, so as to obtain the outline signal of sugarcane based on the occluded elements.

[0015] Furthermore, the calculation method for the binarization process is as follows:

[0016]

[0017] Where f(x) is the original photoelectric signal; g(y) is the binarized signal; x represents all sampling points during the sugarcane scanning process; T R The threshold value is used for the element.

[0018] Furthermore, in step S3, the center frequency and bandwidth of each intrinsic mode function (IMF) in the contour signal are updated iteratively through VMD, so that the contour signal can be adaptively decomposed according to its own characteristics.

[0019] Furthermore, a restricted variable model is constructed through VMD, and the restricted variable model is computed by introducing an enhanced Lagrange multiplication operator to obtain an iterative continuation expression related to the number K of the decomposed intrinsic mode functions (IMFs) and the penalty parameter α.

[0020] Furthermore, in step S3, a fitness function is constructed according to the sparrow search algorithm, and the optimal combination of [K, α] that matches the contour signal is obtained according to the fitness function, so that the contour signal is decomposed into K modal components.

[0021] Furthermore, the fitness function is:

[0022]

[0023]

[0024] Where FuzzyEn is the fuzzy entropy function; d is the random signal time series; m is the embedding dimension; n represents the gradient of the similarity tolerance boundary; and r represents the width of the fuzzy function boundary. and Let r be the probability that two vectors match a real number m or m+1; lg(omega) is the relative clustering algebra of the optimal center frequency; length(w k ) is to extract the optimal center frequency w kThe signal length; P* is the PPMCC of time series x and y; and This represents the average of x and y.

[0025] Furthermore, in step S5, the distance and time information corresponding to the node peaks in the HHT marginal spectrum are obtained through the findpeaks function, and the coordinate position of the sugarcane node is obtained according to the scanning speed in step S1.

[0026] Further, in step S5, a spacing threshold is obtained based on the scanning speed in step S1, and the distance between two adjacent node peaks is compared. When the distance between two adjacent node peaks is greater than the spacing threshold, the positions corresponding to the two node peaks are both sugarcane node positions. When the distance between two adjacent node peaks is less than the spacing threshold, the position corresponding to the node peak with the larger peak value among the two node peaks is the sugarcane node position.

[0027] The beneficial effects of this invention are:

[0028] The vertical projection image of sugarcane is processed using a linear CCD sensor to reflect the corresponding light intensity voltage value of the sugarcane. This eliminates the need for imaging the collected photoelectric signals, reducing hardware costs and the difficulty of subsequent data processing. Furthermore, it avoids physical contact with the sugarcane, minimizing damage to the sugarcane buds. Simultaneously, the original photoelectric signal is binarized with a fixed threshold to reflect the sugarcane outline by the number of occluded image elements, improving processing speed and reducing costs. By decomposing the outline signal into modal components and performing HHT transformation, the signal amplitude distribution with frequency can be accurately reflected, allowing the identification of sugarcane nodes based on the amplitude of the HHT marginal spectrum. Since each sugarcane node region in the HHT marginal spectrum consists of an independent, distinctly convex waveform, the characteristic signal reflects different amplitudes depending on the diameter of the sugarcane node. Therefore, by normalizing the characteristic signal so that its amplitude varies within the range [0,1], a uniform node threshold can be set before node identification, allowing the sugarcane node position to be obtained based on the node threshold. This invention requires only a computer with general performance to improve the efficiency of sugarcane stalk node identification and reduce the cost and time required for non-contact detection of sugarcane stalk nodes. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the efficient sugarcane stem node identification method based on variational mode decomposition of the present invention.

[0030] Figure 2 This is a flowchart of the efficient sugarcane stem node identification method based on variational mode decomposition of the present invention.

[0031] Figure 3This is a parallel light source scanning data image of the efficient sugarcane stem node identification method based on variational mode decomposition of this invention.

[0032] Figure 4 This is a diagram of the original photoelectric signal data of the efficient sugarcane stem node identification method based on variational mode decomposition of this invention.

[0033] Figure 5 This is a binarized data image of the efficient sugarcane stem node identification method based on variational mode decomposition of this invention.

[0034] Figure 6 This is an IMF schematic diagram of the efficient sugarcane stem node identification method based on variational mode decomposition of the present invention.

[0035] Figure 7 This is a schematic diagram of the HHT marginal spectrum of the efficient sugarcane stem node identification method based on variational mode decomposition of this invention.

[0036] Figure 8 This is a schematic diagram of the experimental results of the efficient sugarcane stem node identification method based on variational mode decomposition of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0038] Please refer to Figure 1-2 As shown, a preferred embodiment of the present invention provides an efficient sugarcane stem node identification method based on variational mode decomposition, comprising the following steps:

[0039] S1. A vertical projection image of the sugarcane is obtained by scanning with a parallel light source parallel to the radial direction of the sugarcane, and the vertical projection image is used to obtain the original photoelectric signal through a linear CCD sensor.

[0040] S2. The original photoelectric signal from step S1 is binarized with a fixed threshold to obtain the outline signal of the sugarcane.

[0041] S3. The contour signal obtained in step S2 is decomposed into several modal components using the VMD algorithm, and HHT transformation is performed on each modal component to obtain the HHT marginal spectrum corresponding to each modal component.

[0042] S4. Select the HHT marginal spectrum of the maximum energy range of the low-frequency part of the signal in step S3 as the characteristic signal of the sugarcane node, and normalize the characteristic signal to set the node threshold according to the saturation value of the normalized characteristic signal.

[0043] S5. Obtain the peaks in the normalized feature signal that are greater than the node threshold in step S4 to obtain the node peaks, and take the positions corresponding to the node peaks as the sugarcane node positions.

[0044] The vertical projection image of sugarcane is processed using a linear CCD sensor to reflect the corresponding light intensity voltage value of the sugarcane. This eliminates the need for imaging the collected photoelectric signals, reducing hardware costs and simplifying subsequent data processing. Furthermore, it avoids physical contact with the sugarcane, minimizing damage to the sugarcane buds. Simultaneously, the original photoelectric signal is binarized with a fixed threshold to reflect the sugarcane outline by the number of occluded image elements, improving processing speed and reducing costs. By decomposing the outline signal into modal components and performing HHT transformation, the signal amplitude distribution with frequency can be accurately reflected, allowing for the identification of sugarcane node characteristics based on the amplitude of the HHT marginal spectrum. Since each sugarcane node region in the HHT marginal spectrum consists of an independent, distinctly convex waveform, the characteristic signal reflects different amplitudes depending on the diameter of the sugarcane node. Therefore, by normalizing the characteristic signal so that its amplitude varies within the range [0, 1], a uniform node threshold can be set before node identification, allowing the sugarcane node position to be obtained based on the node threshold. This invention requires only a computer with general performance to improve the efficiency of sugarcane stalk node identification and reduce the cost and time required for non-contact detection of sugarcane stalk nodes.

[0045] In step S1, the data obtained by parallel light source scanning is as follows: Figure 3 As shown, the original photoelectric signal data is as follows: Figure 4 As shown.

[0046] In step S2, element thresholds are obtained through Otsu threshold selection, and the original photoelectric signal is preprocessed according to binarization to obtain the occluded elements in each element threshold, so as to obtain the outline signal of sugarcane based on the occluded elements.

[0047] The original photoelectric signal can be considered as a 1×N grayscale image. Assuming there are two distributions in the image, pixels can be grouped into two classes using an element threshold t. Let's consider an image with L grayscale levels, where the intensity range is set to [tmin, tmax]. The inter-class variance with respect to t is... As shown below:

[0048]

[0049] Where ω0(t) and ω1(t) are the cumulative probabilities, u0(t) and u1(t) are the average values ​​of the two classes of pixels, respectively. T This is the average of the entire dataset. Maximize as shown below:

[0050]

[0051] In this embodiment, in every 5000 elements of photoelectric signal, elements with voltage values ​​below a threshold are considered to be occluded by the projection, while elements with voltage values ​​above the threshold are considered not occluded. The length information of the sugarcane projection can be reflected by the total number of occluded elements. Since the elements on both sides of the CCD provide correction, the data they acquire does not reflect the actual illumination conditions, but this does not significantly affect the trend of the generated contour signal. Therefore, the calculation method for binarization is as follows:

[0052]

[0053] Where f(x) is the original photoelectric signal; g(y) is the binarized signal; x represents all sampling points during the sugarcane scanning process; T R The element threshold is used. A fixed threshold obtained through Otsu threshold selection is used to preprocess the binarization, converting the projection information into the number of occluded elements to reflect the contour information of the sugarcane, such as... Figure 5 As shown.

[0054] In step S3, the center frequency and bandwidth of each intrinsic mode function (IMF) in the contour signal are updated iteratively through VMD, so that the contour signal can be adaptively decomposed according to its own characteristics.

[0055] A restricted variable model is constructed using VMD, and the restricted variable model is computed by introducing an enhanced Lagrange multiplication operator to obtain an iterative expression related to the number K of the decomposed intrinsic mode functions (IMFs) and the penalty parameter α, which continues until convergence.

[0056] VMD is a non-recursive, adaptive, quasi-orthogonal signal decomposition method with significant advantages in processing nonlinear and non-stationary signals. This method is highly efficient, accurately separates signals, and can achieve good noise filtering by utilizing its own Wiener filtering function. Its core concept is to update the center frequency and bandwidth of each intrinsic mode function (IMF) through an iterative method, and adaptively decompose the signal according to the characteristics of the signal itself.

[0057] IMF can be represented as an AM-FM signal, as shown below:

[0058]

[0059] Among them, envelope A k (t) is u k The instantaneous amplitude of (t), It is a non-decreasing instantaneous phase function, so the instantaneous frequency is...

[0060] Bandwidth is obtained by the square of the gradient L 2 The criteria are used for estimation. Therefore, the VMD process can be considered as constructing and solving a restricted variable problem, described as:

[0061]

[0062] Where f is the original input signal, {u k} represents the center frequency of the Kth IMF after decomposition, and δ(t) is the Dirac function.

[0063] To solve formula (4.1), an enhanced Lagrange multiplication operator of the following form is introduced:

[0064]

[0065] Where α is the quadratic penalty, which guarantees the accuracy of signal reconstruction in the presence of Gaussian noise; λ is the Lagrange multiplier, which strictly enforces the constraints.

[0066] Alternating Multiplication Method (ADMM) is used to find the saddle point of Equation (4.2) corresponding to the solution of Equation (4.1). First, the number of decomposition modes is specified beforehand, and the frequency domain expression of the modes is initialized. corresponding center frequency and Lagrange multipliers Subsequently, the pattern and center frequency w k Updated from formulas (4.3) and (4.4):

[0067]

[0068]

[0069] After each update, the pattern and center frequency are obtained. The Lagrange multiplier is also updated according to the following formula:

[0070]

[0071] The above iterations continue until convergence, that is,

[0072]

[0073] From the above description, four parameters need to be manually specified beforehand: K, α, τandε. Compared to the first two parameters, τandε has little impact on the decomposition results, so the default values ​​from the original VMD algorithm are usually used. However, when the value of K is not chosen properly, insufficient or excessive decomposition of the binarized signal may occur. α is related to the suppression performance of noise interference and should also be chosen carefully. Therefore, finding the optimal combination of parameters [K, α] that matches the sugarcane contour signal to be analyzed is crucial for the VMD algorithm.

[0074] In step S3, a fitness function is constructed according to the sparrow search algorithm, and the optimal combination of [K, α] that matches the contour signal is obtained according to the fitness function, so that the contour signal is decomposed into K modal components.

[0075] The Sparrow Search Algorithm (SSA) is a novel heuristic algorithm based on the foraging and anti-predation behaviors of sparrow populations. It divides sparrow populations into producers and foragers, and calculates and updates population positions by constructing a mathematical model. By applying SSA to a combination of VMD parameters, it seeks the optimal value of K and adapts to the characteristics and complexity of the signal.

[0076] The producer's position has been updated:

[0077]

[0078] Where m represents the current iteration; This represents the position of the i-th sparrow in the j-th dimension; α∈(0,1) is a random number; iter max R is a constant with the highest number of iterations; R2∈[0,1] and ST∈[0.5,1.0] represent the alarm value and safety threshold, respectively; Q is a random number following a normal distribution; L is a 1×d matrix with each element having a value of 1. When R2<ST, it means there are no predators nearby, and the producer enters a broad search mode. When R2≥ST, some sparrows have discovered predators, and all sparrows need to quickly fly to other safe areas.

[0079] The location of the freeloader has been updated:

[0080]

[0081] Where c is the number of sparrows, and Xp is the optimal position of the producer; X worst This represents the current worst-case position globally; A represents a 1×d matrix where each element is randomly assigned a value of 1 or -1, and the pseudo-inverse matrix... for The i-th sparrow with the worst fitness value is the most likely to be hungry.

[0082] The number of sparrows aware of danger is generally 10-20% of the total population, and their initial location is randomly generated throughout the population. This mathematical model can be represented as:

[0083]

[0084] Among them, X best The current globally optimal position is considered safe; β is the step size control parameter, a normally distributed random number with a mean of 0 and a variance of 1; K∈[-1, 1] is also a random number; ε is the minimum constant to avoid a denominator of 0. i This represents the sparrow's current fitness value, indicating the best and worst fitness values ​​globally. When f... i <f g At this time, sparrows on the periphery of the population are more vulnerable to predators; if f i =f g At that time, the sparrows in the middle of the group realized the danger and needed to move closer to other sparrows to reduce the risk of being preyed upon.

[0085] The parameter-adaptive VMD method based on SSA iteratively searches for the optimal combination of VMD parameters [K] as the position of the sparrow population. Considering three evaluation metrics—fuzzy entropy, clustering algebra, and Pearson correlation coefficient (PPMCC)—a fitness function optimized by SSA is constructed:

[0086]

[0087] in,

[0088]

[0089] In formulas (2) and (3), FuzzyEn is the fuzzy entropy function; d is the random signal time series; m is the embedding dimension; n represents the gradient of the similarity tolerance boundary; and r represents the width of the fuzzy function boundary. and Let r be the probability that two vectors match a real number m or m+1; lg(omega) is the relative clustering algebra of the optimal center frequency; length(w k ) is to extract the optimal center frequency w k The signal length; P* is the PPMCC of time series x and y; and This represents the average of x and y.

[0090] In step S3 of this embodiment, the minimum fitness value of SSA, 0.04, appears in the fifth iteration, at which point α... best For 1798, K bestThe optimal value for K is 6. Therefore, in the next section, the VMD algorithm is initialized using the parameters in Table 3 to decompose the binarized signal to extract the feature signal. Except for τ and ε using default parameters, DC is set to 0, disregarding the DC component, and init is set to 1 to uniformly initialize omega. The optimal value of K is obtained as 6 through the fitness function, thus yielding six IMFs to decompose the contour signal into K modal components, such as... Figure 6 As shown.

[0091] In step S3, the HHT marginal spectrum accurately reflects the distribution of signal amplitude with frequency, and its amplitude represents the probability of that frequency occurring. By using the characteristics of the HHT marginal spectrum to identify the features of sugarcane nodes, the HHT marginal spectrum corresponding to the modal components is obtained, such as... Figure 7 As shown.

[0092] In step S4, according to Figure 7 As shown, IMF1 is the dominant frequency band, with the highest energy distribution in the low-frequency range of 0-5Hz, representing the overall diameter variation of sugarcane along the scanning direction. IMF2, IMF3, and IMF4 have relatively high amplitudes and narrow bandwidths, representing the mid-to-low frequency part of the signal, where the characteristic energy of sugarcane is mainly concentrated. IMF2 indicates the location of the root zone, while IMF3 and IMF4 represent detailed information about the growth ring, wax ring, and buds within the root zone. IMF5 and IMF6 have smaller amplitudes and wider bandwidths. The dark current of the CCD sensor, interference during transmission, and other noise energy account for a large proportion of the high-frequency part of the signal. The mode mixing phenomenon between them and IMF4 is the slightest, so they are considered irrelevant modes for removal when selecting characteristic signals.

[0093] Since the focus of this embodiment is on the regional localization of sugarcane nodes, although IMF1 has the highest probability of appearing in the contour signal, its influence on the signal trend is not conducive to setting the threshold for judging nodes. IMF3 and IMF4 contain too much detailed information to accurately locate nodes. Therefore, in the next section, IMF2, with a maximum energy range of 7-12 Hz, is selected as the characteristic signal for determining and locating sugarcane nodes.

[0094] In step S4, according to Figure 7 As shown, each sugarcane node region consists of an independent, distinct upward-convex waveform. The characteristic signal reflects different amplitudes depending on the diameter of the sugarcane node. By normalizing the characteristic signal so that its amplitude varies within the range of [0,1], it is easier to determine the node threshold. In this embodiment, the node threshold is 85% of the saturation value of the normalized characteristic signal.

[0095] In step S5, the distance and time information corresponding to the node peaks in the HHT marginal spectrum are obtained through the findpeaks function, and the coordinate position of the sugarcane node is obtained according to the scanning speed in step S1.

[0096] In step S5, a spacing threshold is obtained based on the scanning speed in step S1. The distance between two adjacent node peaks is compared. When the distance between two adjacent node peaks is greater than the spacing threshold, the positions corresponding to the two node peaks are both sugarcane node positions. When the distance between two adjacent node peaks is less than the spacing threshold, the position corresponding to the node peak with the larger peak value is the sugarcane node position.

[0097] In this embodiment, the coordinates of the sugarcane nodes can be calculated using MATLAB software, where T N The threshold for determining sugarcane nodes is determined by storing the distance and time information corresponding to all peaks in the characteristic signal calculated by the findpeaks function (a function included in MATLAB) in Apks and Alocs; the time information of the peaks determined to be sugarcane nodes is stored in Blocs; and the coordinates of each sugarcane node are calculated by the velocity VC set by the photoelectric scanning unit and stored in S.

[0098] The efficient sugarcane stem node identification method based on variational mode decomposition in this embodiment is compared with manual measurement to verify the effectiveness of this embodiment. Let the node position identified by the algorithm be Si, the actual position of the stem measured manually be Xi, and the number of nodes identified on the sugarcane be n. Then, the identification error of the sugarcane is calculated by E:

[0099]

[0100] In this embodiment, a recognition error E greater than twice the average stem length (i.e., 45.42 mm) is considered a false recognition, while a recognition error E less than one-third of the average node length (i.e., 7.57 mm) is considered a correct recognition. Therefore, the recognition rate can be calculated as follows:

[0101]

[0102] Where, N T N is the total number of nodes within the scanning range of the photoelectric detection unit for all experimental samples. C This represents the number of nodes correctly identified by the algorithm. In evaluating the algorithm's response time, in this embodiment, the total time from when the algorithm is invoked by the PC to when the sugarcane is scanned once by the photoelectric detection unit and the recognition result is returned is defined as the algorithm's response time.

[0103] The experimental results after individually testing all 32 sugarcane samples at a scanning speed of 75 mm / s and an illuminance of 91.91 lx are shown below. Figure 8 It can be observed that the overall recognition rate of sugarcane nodes exceeds 95%, the average execution time of the algorithm is 0.13 seconds, and the average absolute error of the algorithm in locating nodes is less than one-third of the average length of sugarcane nodes (i.e., 7.57 mm).

[0104] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0105] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A highly efficient sugarcane stem node identification method based on variational mode decomposition, characterized in that, The method for identifying sugarcane stem nodes includes the following steps: S1. A vertical projection image of the sugarcane is obtained by scanning with a parallel light source parallel to the radial direction of the sugarcane, and the vertical projection image is used to obtain the original photoelectric signal through a linear CCD sensor. S2. Binarize the original photoelectric signal from step S1 with a fixed threshold to obtain the outline signal of the sugarcane. S3. The contour signal obtained in step S2 is decomposed into several modal components by the VMD algorithm, and HHT transformation is performed on each modal component to obtain the HHT marginal spectrum corresponding to each modal component. In step S3, the center frequency and bandwidth of each intrinsic mode function (IMF) in the contour signal are updated iteratively through VMD, so that the contour signal can be adaptively decomposed according to its own characteristics. A restricted variable model is constructed by VMD, and the restricted variable model is computed by introducing an enhanced Lagrange multiplication operator to obtain an iterative expression related to the number K of the decomposed intrinsic mode functions (IMFs) and the penalty parameter α, which continues until convergence. In step S3, a fitness function is constructed based on the sparrow search algorithm, and [K,] a value matching the contour signal is obtained based on the fitness function. The optimal combination is used to decompose the contour signal into K modal components; S4. Select the HHT marginal spectrum of the maximum energy range of the low-frequency part of the signal in step S3 as the characteristic signal of the sugarcane node, and normalize the characteristic signal to set the node threshold according to the saturation value of the normalized characteristic signal. S5. Obtain the peaks in the normalized feature signal that are greater than the node threshold in step S4 to obtain the node peaks, and take the positions corresponding to the node peaks as the sugarcane node positions.

2. The efficient sugarcane stem node identification method based on variational mode decomposition according to claim 1, characterized in that: In step S2, element thresholds are obtained through Otsu threshold selection, and the original photoelectric signal is preprocessed according to binarization to obtain the occluded elements in each element threshold, so as to obtain the outline signal of sugarcane based on the occluded elements.

3. The efficient sugarcane stem node identification method based on variational mode decomposition according to claim 2, characterized in that: The calculation method for the binarization process is as follows: Official (1) Where f(x) is the original photoelectric signal; g(y) is the binarized signal; x represents all sampling points during the sugarcane scanning process; T R This is the threshold value for the element.

4. The efficient sugarcane stem node identification method based on variational mode decomposition according to claim 1, characterized in that: The fitness function is: Official (2) Official (3) Where FuzzyEn is the fuzzy entropy function; d is the random signal time series; m is the embedding dimension; n represents the gradient of the similarity tolerance boundary; and r represents the width of the fuzzy function boundary. and Let r be the probability that two vectors match a real number m or m+1; lg(omega) is the relative clustering algebra of the optimal center frequency; length(w k ) is to extract the optimal center frequency w k The signal length; P* is the PPMCC of time series x and y; and This represents the average of x and y.

5. The efficient sugarcane stem node identification method based on variational mode decomposition according to claim 1, characterized in that: In step S5, the distance and time information corresponding to the node peaks in the HHT marginal spectrum are obtained through the findpeaks function, and the coordinate positions of the sugarcane nodes are obtained according to the scanning speed in step S1.

6. The efficient sugarcane stem node identification method based on variational mode decomposition according to claim 1, characterized in that: In step S5, a spacing threshold is obtained based on the scanning speed in step S1. The distance between two adjacent node peaks is compared. When the distance between two adjacent node peaks is greater than the spacing threshold, the positions corresponding to the two node peaks are both sugarcane node positions. When the distance between two adjacent node peaks is less than the spacing threshold, the position corresponding to the node peak with the larger peak value is the sugarcane node position.

Images