Intelligent identification and calculation method for growth phenotypic characteristics of edible fungi in germination period

Abstract

The invention discloses an intelligent identification and calculation method for growth phenotypic characteristics of edible fungi in a budding period, and relates to the technical field of modern facility agriculture. Comprising the following steps: shooting and obtaining a clear RGB image of field growth in the edible mushroom germination period; performing contrast enhancement processing on the clear RGB image; accurate identification is carried out through an improved instance segmentation model RSL-YOLOv8m-Seg to obtain a bud primordium mask image; traversing and segmenting each bud primordium mask one by one, and extracting the contour of the mask; reconstructing a bud primordium contour by adopting a contour fitting method based on each mask contour; extracting morphological characteristics of the bud primordium; and counting growth phenotypic characteristics in a budding period. According to the method, the growth characteristics of the edible fungi in the germination period can be accurately calculated, and a phenotypic characteristic basis is provided for intelligent and accurate management and control of an edible fungi house.

Description

Technical Field

[0001] This invention relates to the field of modern facility agriculture technology, and in particular to an intelligent recognition and calculation method for the growth phenotypic characteristics of edible fungi during the budding stage. Background Technology

[0002] In recent years, the industrialized production model of edible fungi has made significant progress, with a high level of automation in the entire production process. However, the level of intelligence in mushroom house cultivation technology, a key aspect of edible fungi production, remains low, urgently requiring intelligent and precise control to improve the yield and quality of edible fungi. The budding stage is a critical phase in fruiting management, as its growth status directly determines the quantity and quality of subsequent fruiting bodies, thus directly affecting yield and quality. However, most companies still rely on manual methods to inspect crop growth and adjust environmental parameters such as temperature, humidity, ventilation, and light based on experience. Furthermore, growth changes during the budding stage are subtle and their characteristics are difficult to detect; manual observation is prone to errors and delays, making it difficult to meet the needs of refined management.

[0003] With the expansion of edible mushroom production scale and the increasing complexity of factory cultivation models, extracting key growth characteristics during the budding stage has become a technical challenge. Currently, there is a lack of research on intelligent identification of bud primordia and intelligent computation of their phenotypic features. The small overall size of bud primordia, their significant multi-scale characteristics, low contrast with the culture medium, and blurred edges all increase the difficulty of identification. Furthermore, bud primordia tend to grow at high density, and target adhesion is very common, making traditional instance segmentation algorithms based on regular shape assumptions perform poorly in this scenario. In addition, the unique irregular shape, mutual occlusion, and dynamic growth characteristics of bud primordia make conventional geometric feature extraction methods insufficient to meet accuracy requirements. The combination of these complex factors results in significant deficiencies in identification accuracy, stability, and applicability, urgently requiring the development of innovative intelligent solutions.

[0004] Therefore, proposing an intelligent recognition and calculation method for the growth phenotypic characteristics of edible fungi during the budding stage to solve the difficulties existing in the prior art is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent identification and calculation method for the growth phenotypic characteristics of edible fungi during the budding stage. This method can automatically identify and calculate the growth characteristics of edible fungi during the budding stage, providing phenotypic characteristic data for the intelligent management of the cultivation environment during the budding stage of industrialized edible fungi production. This helps to improve the yield and quality of industrialized edible fungi and effectively promotes the intelligent production of edible fungi facility agriculture.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for intelligently identifying and calculating the phenotypic characteristics of edible fungi during the budding stage, comprising:

[0008] S1. Capture and obtain clear RGB growth images of edible fungi during the sprouting stage;

[0009] S2. Enhance the contrast of the clear RGB grown image;

[0010] S3. Accurately identify and obtain the bud primordium mask through instance segmentation model;

[0011] S4. Iterate through each bud primordium mask and extract the outline of each mask;

[0012] S5. Based on the contours of each mask, the contour fitting method is used to reconstruct the bud primordium contour;

[0013] S6. Extract the morphological characteristics of the bud primordia;

[0014] S7. Statistical analysis of growth phenotypes during the budding stage.

[0015] Optionally, in S2, the clear RGB grown image undergoes contrast enhancement processing using the contrast-limited adaptive histogram equalization (CLAHE) algorithm, specifically:

[0016] S201. Convert the clear RGB grown image into a single-channel grayscale image. gray ;

[0017] S202, Transfer grayscale image I gray Divide into several non-overlapping sub-blocks B of size M×N. i,j ,in H and W represent the height and width of the image, respectively.

[0018] S203, For each sub-block B i，j Calculate the local histogram h i,j (k), based on the cutting limit T clip Crop the histogram:

[0019]

[0020] Where k∈[0,L-1] is the gray level, and L is the maximum gray level;

[0021] S204. Distribute the cropped excess pixels evenly across all gray levels to obtain the adjusted histogram h. i,j ''(k), for h i,j ''(k) performs cumulative distribution function (CDF) calculation and generates grayscale mapping function f i,j (k);

[0022] S205. For each pixel (x, y) in the image, according to the mapping function f of its four adjacent sub-blocks... i,j f i,j+1 f i+1,j f i+1,j+1 Perform bilinear interpolation to obtain the final grayscale value I. enhanced (x,y):

[0023]

[0024] Among them, W m,n These are the interpolation weighting coefficients;

[0025] S206, Output Enhanced Image I enhanced ;

[0026] The above method, optionally, W m,n The method for obtaining it is as follows:

[0027]

[0028] in And W 00 +W 01 +W 10 +W 11 =1.

[0029] Optionally, in S3, the instance segmentation model used in the above method is the improved YOLOv8 model - RSL-YOLOv8m-Seg, specifically:

[0030] S301. Using the basic YOLOv8-seg structure, replace the Conv module with the RFCAConv module, introduce the SKAttention module between the C2f module and the SPPF module, and replace the SPPF module with the SPPF_LSKA module.

[0031] S302. The input image is sequentially downsampled through the improved RFCAConv and C2f modules to generate multi-level feature maps from P1 to P5. The SKAttention module is introduced into the deep network to realize dynamic channel weight allocation, and the spatial features of different receptive fields are fused through the SPPF_LSKA module to finally output 1024-dimensional high-level features containing details and semantic information.

[0032] The RFCAConv module is a convolutional unit that introduces receptive field channel attention, and its expression is:

[0033]

[0034] Where R(·) represents multi-branch receptive field extraction, CA(·) is channel attention operation, and σ is the Sigmoid activation function;

[0035] The SKAttention module is an attention mechanism that dynamically selects multi-scale receptive fields, and its processing satisfies the following:

[0036]

[0037] Among them, DepthwiseConv 3×3 For depthwise separable convolutions, GAP(·) denotes global average pooling. The MLP contains fully connected layers with a compression ratio of r, where W∈R. C×2 This is a dynamic branch weight matrix;

[0038] The SPPF_LSKA module employs a large-kernel spatial attention mechanism, and its pooling process satisfies the following:

[0039]

[0040] Among them, LSKA k MaxPool is a learnable k×k spatial attention kernel whose weights are dynamically generated through spatial offset convolution. k This represents k×k max pooling;

[0041] S303. Upsample the P5 feature and concatenate it with the P4 feature. After optimization by the C2f module, a 512-dimensional mesoscale feature is generated. Upsample and fuse the P3 feature again to output a 256-dimensional low-level feature map containing rich details.

[0042] S304. After downsampling the P3 level features, they are concatenated with the P4 features to generate 512-dimensional mid-level features; further downsampling and fusion with the P5 features outputs 1024-dimensional high-level features with enhanced semantics, forming a bidirectional feature pyramid structure.

[0043] S305, by jointly utilizing the three-level features of P3, P4 and P5, and through dual-scale masking, parallel prediction of pixel-level segmentation results and target detection boxes, end-to-end instance segmentation is achieved.

[0044] S306, Output bud primordium mask.

[0045] Optionally, in S5, the contour fitting method used in the above method is convex hull contour fitting, specifically:

[0046] Extracting the contour point set Calculate the convex hull H = ConvexHull(C) of the contour point set, satisfying:

[0047]

[0048] Where K≤N is the number of vertices of the convex hull.

[0049] In the above method, optionally, in S6, calculating the morphological characteristics of the bud primordium is equivalent to calculating the height-to-width ratio of the bud primordium, specifically as follows:

[0050] S601, Extract the vertex set of the convex hull;

[0051] S602, Search for the two points (P) that are farthest apart from each other in the convex hull vertex set. max1 ,P max2 ), forming the main height axis L h :

[0052]

[0053] S603, Calculate L h vertical vector Where [Δx,Δy]=P max1 -P max2 ;

[0054] S604, along V ⊥ Projecting convex hull vertices in the direction determines the pair of points with the maximum width (W). start W end ):

[0055]

[0056] S605, Calculate the morphological characteristics of the target

[0057] The above method, optionally, includes the following phenotypic characteristics of edible fungi during the budding stage in S7: basic bud quantity and effective bud quantity.

[0058] Optionally, the basic number of buds can be calculated using the total number of bud primordia masks N obtained from the above method. base express.

[0059] The above method, optionally, involves calculating the effective number of buds as follows: statistically analyzing the morphological characteristics of all bud primordia, and then calculating based on the morphological characteristics R of the primordia. aspect Screening for the effective number of buds N that meets the requirements use :

[0060]

[0061] The value of n depends on the specific growth standards of the edible fungi.

[0062] As can be seen from the above technical solution, the present invention provides an intelligent recognition and calculation method for the growth phenotypic characteristics of edible fungi during the budding stage, which has the following beneficial effects:

[0063] 1) This invention, for the first time in a factory-scale production and cultivation environment, realizes intelligent identification and calculation of the growth phenotypic characteristics of edible fungi during the budding stage;

[0064] 2) The CLAHE algorithm was used to process the growth images of edible fungi bud primordia. The CLAHE-processed images of bud primordia showed a more balanced brightness distribution and clearer contours, effectively enhancing detail information. Furthermore, it significantly reduced detail loss caused by uneven illumination, optimized the local contrast between the bud primordia and the complex background, thereby improving the recognizability of the bud primordia and achieving more accurate segmentation of the bud primordia body from the complex background.

[0065] 3) To address the challenges of bud primordia detection involving multiple scales, dense distribution, and overlapping occlusions, the proposed improved YOLOv8 algorithm RSL-YOLOv8m-Seg introduces the RFCAConv module, SKAttention module, and SPPF_LSKA module. The RFCAConv module extracts and enhances small target features through multi-branch receptive fields, the SKAttention module adaptively fuses spatial features of different scales to solve the problem of dense occlusion, and SPPF_LSKA optimizes contour localization through a large-kernel spatial attention mechanism. The synergistic effect of the three modules effectively improves the segmentation effect and accuracy of bud primordia at multiple scales.

[0066] 4) Convex hull contour fitting is used to reconstruct the contour of the bud primordium. By constructing the minimum convex polygon, the mask shape can be simplified and the influence of irregular shape can be reduced, so that the reconstructed contour is closer to the actual contour of the bud primordium. Based on the closure property of the convex hull, the reasonable boundary of the occluded area can also be better inferred. This effectively solves the problem of large deviation in the reconstructed contour caused by the missing part of the entity of the occluded bud primordium, and greatly improves the accuracy of edge contour fitting of densely occluded complex bud primordium.

[0067] 5) This invention proposes for the first time a method for calculating the morphological characteristics of bud primordia based on convex hull vertex sets and a method for determining the effective number of buds. This invention provides ideas and methods for the intelligent identification and calculation of the phenotypic characteristics of edible fungi bud primordia, which can effectively provide standardized and quantitative evaluation of the growth status during the budding stage, provide a basis for intelligent control decisions of the edible fungi cultivation environment, and can greatly promote the process of intelligent and precise environmental control of edible fungi. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0069] Figure 1 The flowchart illustrates a method for intelligent recognition and calculation of growth phenotypic characteristics during the budding stage of edible fungi, provided by this invention.

[0070] Figure 2a To improve the YOLOv8 (RSL-YOLOv8m-Seg) model for bud primordia recognition.

[0071] Figure 2b Based on the YOLOv8 model.

[0072] Figure 3 Image of bud primordia of edible fungi.

[0073] Figure 4 for Figure 3 The result image after processing with CLAHE.

[0074] Figure 5a The result image for recognition by the improved YOLOv8 instance segmentation model (RSL-YOLOv8m-Seg).

[0075] Figure 5b The result of the YOLOv8 instance segmentation model identification.

[0076] Figure 6 for Figure 5a Image after reconstruction of the bud primordium mask contour.

[0077] Figure 7 for Figure 6 The contours are reconstructed using different contour fitting methods.

[0078] Figure 8a An inaccurate bud primordium contour image fitted to a rotated rectangular contour.

[0079] Figure 8b An inaccurate bud primordium contour image after fitting an elliptical contour.

[0080] Figure 8c This is the bud primordium contour image after fitting the convex hull contour.

[0081] Figure 9 The number of bud primordia within each morphological region.

[0082] Figure 10 Visualization of the number of effective buds selected. Detailed Implementation

[0083] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0085] Reference Figure 1 As shown, this invention discloses an intelligent recognition and calculation method for the growth phenotypic characteristics of edible fungi during the budding stage, including:

[0086] S1. Capture and obtain clear RGB images of the budding stage of edible fungi grown in bottles and bags;

[0087] S2. Enhance the contrast of clear RGB images;

[0088] S3. Accurately identify and obtain the bud primordium mask image through the instance segmentation model;

[0089] S4. Iterate through each bud primordium mask and extract the outline of each mask;

[0090] S5. Based on the contours of each mask, the contour fitting method is used to reconstruct the bud primordium contour;

[0091] S6. Extract the morphological characteristics of the bud primordia;

[0092] S7. Statistical analysis of growth phenotype characteristics during the budding stage in a single culture bottle.

[0093] Furthermore, in S2, the clear RGB image undergoes contrast enhancement processing using the CLAHE algorithm, specifically:

[0094] S201. Convert a clear RGB image into a single-channel grayscale image. gray ;

[0095] S202, Transfer grayscale image I gray Divide into several non-overlapping sub-blocks B of size M×N. i,j ,in H and W represent the height and width of the image, respectively.

[0096] S203, For each sub-block B i，j Calculate the local histogram h i,j (k), based on the cutting limit T clip Crop the histogram:

[0097]

[0098] Where k∈[0,L-1] is the gray level, and L is the maximum gray level;

[0099] S204. Distribute the cropped excess pixels evenly across all gray levels to obtain the adjusted histogram h. i,j ''(k), for h i,j ''(k) performs cumulative distribution function (CDF) calculation and generates grayscale mapping function f i,j (k);

[0100] S205. For each pixel (x, y) in the image, according to the mapping function f of its four adjacent sub-blocks... i,j f i,j+1 f i+1,j f i+1,j+1 Perform bilinear interpolation to obtain the final grayscale value I. enhanced (x,y):

[0101]

[0102] Among them, W m,n These are the interpolation weighting coefficients;

[0103] S206, Output Enhanced Image I enhanced ;

[0104] The above method, optionally, W m,n The method for obtaining it is as follows:

[0105]

[0106] in And W 00 +W 01 +W 10 +W 11 =1.

[0107] Furthermore, from a visual perspective, the CLAHE-processed images of edible mushroom bud primordia exhibit a more balanced brightness distribution and clearer contours, effectively enhancing detail. It also significantly reduces detail loss caused by uneven lighting, optimizing local contrast and thus improving image recognizability.

[0108] Furthermore, in S3, the instance segmentation model uses the improved YOLOv8 model - RSL-YOLOv8m-Seg, specifically:

[0109] S301. Using the basic YOLOv8-seg structure, replace the Conv module with the RFCAConv module, introduce the SKAttention module between the C2f module and the SPPF module, and replace the SPPF module with the SPPF_LSKA module, as follows. Figure 2a and Figure 2b As shown;

[0110] S302. The input image is sequentially downsampled through the improved RFCAConv and C2f modules to generate multi-level feature maps from P1 to P5. The SKAttention module is introduced into the deep network to realize dynamic channel weight allocation, and the spatial features of different receptive fields are fused through the SPPF_LSKA module to finally output 1024-dimensional high-level features containing details and semantic information.

[0111] The RFCAConv module is a convolutional unit that introduces receptive field channel attention, and its expression is:

[0112]

[0113] Where R(·) represents multi-branch receptive field extraction, CA(·) is channel attention operation, and σ is the Sigmoid activation function;

[0114] The SKAttention module is an attention mechanism that dynamically selects multi-scale receptive fields, and its processing satisfies the following:

[0115]

[0116] Among them, DepthwiseConv 3×3 For depthwise separable convolutions, GAP(·) denotes global average pooling. The MLP contains fully connected layers with a compression ratio of r, where W∈R. C×2 This is a dynamic branch weight matrix;

[0117] The SPPF_LSKA module employs a large-kernel spatial attention mechanism, and its pooling process satisfies the following:

[0118]

[0119] Among them, LSKA k MaxPool is a learnable k×k spatial attention kernel whose weights are dynamically generated through spatial offset convolution. k This represents k×k max pooling;

[0120] S303. Upsample the P5 feature and concatenate it with the P4 feature. After optimization by the C2f module, a 512-dimensional mesoscale feature is generated. Upsample and fuse the P3 feature again to output a 256-dimensional low-level feature map containing rich details.

[0121] S304. After downsampling the P3 level features, they are concatenated with the P4 features to generate 512-dimensional mid-level features; further downsampling and fusion with the P5 features outputs 1024-dimensional high-level features with enhanced semantics, forming a bidirectional feature pyramid structure.

[0122] S305, by jointly utilizing the three-level features of P3, P4 and P5, and through dual-scale masking, parallel prediction of pixel-level segmentation results and target detection boxes, end-to-end instance segmentation is achieved.

[0123] S306, Output bud primordium mask.

[0124] Furthermore, compared to the YOLOv8m instance segmentation model, the three core modules introduced achieve performance improvements through multi-dimensional feature optimization: the RFCAConv module enhances the feature representation ability of small targets through multi-scale receptive field extraction and channel attention mechanisms; the SKAttention module adopts a dynamic convolution kernel selection strategy to adaptively fuse spatial features of different scales, effectively improving the recognition effect of overlapping target regions; and the SPPF_LSKA module introduces a large-kernel spatial attention mechanism to establish long-range spatial dependencies in a multi-level feature pyramid, improving the target contour restoration ability under occlusion conditions. These three modules complement each other from the perspectives of channel dimension, spatial scale, and receptive field range, jointly solving the feature extraction and segmentation challenges in scenarios with densely distributed small targets.

[0125] Furthermore, in S5, the bud primordium contour is reconstructed based on the contour fitting method of each mask contour;

[0126] Extracting the contour point set Calculate the convex hull H = ConvexHull(C) of the contour point set, satisfying:

[0127]

[0128] Where K≤N is the number of vertices of the convex hull;

[0129] Furthermore, in S6, the morphological characteristics of the bud primordium are calculated as follows:

[0130] S601, Extract the vertex set of the convex hull;

[0131] S602, Search for the two points (P) that are farthest apart from each other in the convex hull vertex set. max1 ,P max2 ), forming the main height axis L h :

[0132]

[0133] S603, Calculate L h vertical vector Where [Δx,Δy]=P max1 -P max2 ;

[0134] S604, along V ⊥ Projecting convex hull vertices in the direction determines the pair of points with the maximum width (W). start W end ):

[0135]

[0136] S605, Calculate the morphological characteristics of the target

[0137] Furthermore, in S7, the phenotypic characteristics of edible fungi during the budding stage are: basic bud quantity and effective bud quantity.

[0138] Furthermore, the basic bud quantity is calculated using the total number of bud primordia masks N extracted from the segmentation. base express.

[0139] Furthermore, the method for calculating the effective number of buds is as follows: statistically analyze the morphological characteristics of all bud primordia, and based on the morphological characteristics R of the primordia... aspect Screening for the effective number of buds N that meets the requirements use :

[0140]

[0141] The value of n depends on the specific growth standards of the edible fungi.

[0142] In one specific embodiment, refer to Figure 3 As shown, images of edible fungi during their sprouting stage were captured and obtained.

[0143] Convert the growth images of edible fungi during the budding stage from RGB images to grayscale images;

[0144] To perform CLAHE contrast enhancement, the image is first converted to grayscale and divided into several non-overlapping sub-blocks. A local histogram is calculated for each sub-block, and cropping and limiting are applied. Excess pixels are then evenly distributed across different grayscale levels. Next, a cumulative distribution function is calculated based on the adjusted histogram, generating a grayscale mapping function. Finally, bilinear interpolation is used to integrate the mapping relationships between adjacent sub-blocks, outputting the enhanced grayscale image, as shown below. Figure 4 As shown.

[0145] Recognition is performed using the improved YOLOv8 instance segmentation model RSL-YOLOv8m-Seg: the input image is downsampled stepwise through the RFCAConv and C2f modules to generate P1-P5 multi-scale features. The deep features are enhanced by the SKAttention and SPPF_LSKA modules, and a P3-P5 feature pyramid containing detailed and semantic information is constructed through bidirectional feature fusion. Finally, the detection boxes for bud primordia and pixel-level segmentation results are output by the dual-scale masking mechanism, such as... Figure 5a As shown.

[0146] The results of YOLOv8-seg instance segmentation model identification are as follows: Figure 5b As shown, due to the small size, dense growth distribution, and overlapping occlusion of the bud primordia, the recognition rate of bud primordia is low, with only 638 bud primordia identified. In contrast, the method proposed in this invention identifies 765 bud primordia, effectively reducing the impact of small size, dense distribution, and mutual occlusion of bud primordia. RSL-YOLOv8m-Seg exhibits high segmentation accuracy, with an F1 score of 54.46%, a P-value of 78.5%, and an R-value of 41.7%, all higher than other typical instance segmentation algorithms, as shown in Table 1. The above demonstrates the significant superiority and innovation of this invention.

[0147] Table 1

[0148]

[0149] The primordia masks of each bud are segmented one by one, and the outline of each mask is extracted, such as... Figure 6 As shown.

[0150] Commonly used contour fitting methods include three types: rotated rectangle fitting, ellipse fitting, and convex hull fitting. The fitting results are as follows: Figure 7 As shown.

[0151] The morphological features of bud primordia are extracted using a rotating rectangle fitting method. This method fits the contour of each bud primordia to a minimum bounding rectangle, ensuring the rectangle tightly encloses the target at any angle, providing a stable directional reference for subsequent geometric parameter calculations. However, this method suffers from inaccurate fitting; the length and width of the fitted rectangle do not accurately represent the aspect ratio of the original contour, and the height of the bud primordia sometimes corresponds to the diagonal of the rectangle. Figure 8a As shown.

[0152] The morphological features of bud primordia are extracted using an ellipse contour fitting method: The ellipse fitting algorithm fits the contour of each bud primordia to an optimally fitted ellipse, allowing the ellipse to approximate the target shape in a symmetrical form. This method is suitable for objects with approximately elliptical features. However, it produces significant shrinkage or expansion errors when dealing with asymmetrical or sharp contours. In such cases, the fitted ellipse contour differs considerably from the original contour, with a large offset between the major axis and the elevation line of the original contour. Figure 8b As shown.

[0153] Morphological features of bud primordia are extracted using the convex hull contour fitting method: the contour of each bud primordia is fitted to a minimum circumscribed convex polygon using the convex hull algorithm, so that the convex hull can effectively surround the external boundary of the bud primordia, eliminating interference caused by noise or irregular shapes in the image. The fitting results are as follows: Figure 8c As shown.

[0154] In practical applications, for targets with irregular triangular pyramidal shapes, such as bud primordia of edible fungi, this invention compared and verified three contour fitting methods: rotating rectangle fitting, which characterizes the target using the minimum bounding rectangle, has high computational efficiency, but produces significant shape distortion for targets with sharp apex; ellipse fitting, which generates symmetrical closed curves based on the least squares method, is prone to contour contraction or expansion deviations at the top of the bud primordia; in contrast, the convex hull fitting algorithm used in this invention, by constructing the minimum convex polygon, can effectively maintain the original geometric features of the target, and its contour fitting degree and geometric parameter calculation accuracy are significantly better than the rotating rectangle and ellipse fitting methods, making it particularly suitable for the morphological analysis of bud primordia with complex edge features.

[0155] Extract the contour point set from the fitted image Calculate the convex hull H = ConvexHull(C) of the contour point set, satisfying:

[0156]

[0157] Where K≤N is the number of vertices of the convex hull;

[0158] Search for the two points (P0, P0) that are farthest apart from each other in the convex hull vertices. max1 ,P max2), forming the main height axis L h :

[0159]

[0160] Calculate L h vertical vector Where [Δx,Δy]=P max1 -P max2 ;

[0161] Along V ⊥ Projecting convex hull vertices in the direction determines the pair of points with the maximum width (W). start W end ):

[0162]

[0163] Calculate target morphological features

[0164] Figure 3 Statistical results of the morphological characteristics of the bud primordia are as follows: Figure 9 As shown.

[0165] Phenotypic characteristics of edible fungi during the budding stage: basic bud quantity and effective bud quantity.

[0166] Among them, the basic bud quantity is calculated using the total number of bud primordia masks N that have been segmented. base The effective number of shoots is calculated as follows: The morphological characteristics of all shoot primordia are statistically analyzed, and the effective number of shoots is determined based on the morphological characteristics R of the primordia. aspect Screening for the effective number of buds N that meets the requirements use .

[0167]

[0168] The value of n is determined according to the growth standards of different types of edible fungi.

[0169] Figure 3 Basic bud quantity N base =45+170+182+142+95+59+34+18+8+8+3+1=765.

[0170] Figure 3 Effective bud count N use =69+34+18+8+3+1=133, the visualization result is as follows Figure 10 As shown.

[0171] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0172] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent recognition and calculation of growth phenotype characteristics of edible mushroom sprout emergence period, characterized in that, The method comprises the following steps: S1, shooting and acquiring clear RGB images of the out-time period of mushroom buds; S2, performing contrast enhancement processing on the clear RGB growth images; S3, accurately identifying and obtaining bud primordium masks through an instance segmentation model; S4, traversing each bud primordium mask and extracting the contour of each mask; S5, reconstructing the bud primordium contour based on the contour of each mask using a contour fitting method; S6, extracting the morphological features of the bud primordium; S7, counting the growth phenotype features of the bud out-time period.

2. The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 1, characterized in that, In S2, the contrast enhancement processing adopts the CLAHE algorithm.

3. The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 1, characterized in that, In S3, the instance segmentation model is an improved YOLOv8 model, RSL-YOLOv8m-Seg, which specifically comprises the following steps: S301, using the basic YOLOv8-seg structure, replacing the Conv module with the RFCAConv module, introducing the SKAttention module between the C2f module and the SPPF module, and replacing the SPPF module with the SPPF_LSKA module; S302, inputting the image into the improved RFCAConv and C2f modules for hierarchical down-sampling to generate P1 to P5 multi-level feature maps; introducing the SKAttention module in the deep network to realize dynamic channel weight allocation, and fusing the spatial features of different receptive fields through the SPPF_LSKA module to finally output 1024-dimensional high-level features containing details and semantic information; The RFCAConv module is a convolution unit that introduces receptive field channel attention, and its expression is as follows: Wherein, R(·) represents multi-branch receptive field extraction CA(·) is channel attention operation, and σ is a Sigmoid activation function; the SKAttention module is an attention mechanism that dynamically selects multi-scale receptive fields, and its processing process satisfies: where DepthwiseConv 3×3 is a depthwise separable convolution, GAP(·) denotes a global average pooling, MLP contains a fully connected layer with compression ratio r, W ∈ R C×2 is a dynamic branch weight matrix; The SPPF_LSKA module adopts a large kernel space attention mechanism, and its pooling process satisfies: where LSKA k is a learnable k x k spatial attention kernel whose weights are dynamically generated by spatial offset convolution, MaxPool k denotes k x k max pooling; S303, up-sampling the P5 feature and splicing it with the P4 feature, generating 512-dimensional medium-scale features after optimization by the C2f module; again up-sampling and fusing the P3 feature, outputting a 256-dimensional low-level feature map containing rich details; S304, down-sampling the P3-level feature and splicing it with the P4 feature, optimizing to generate 512-dimensional medium-level features; further down-sampling and fusing the P5 feature, outputting 1024-dimensional high-level features with enhanced semantics, forming a bidirectional feature pyramid structure; S305, jointly using the P3, P4, and P5 three-level features, predicting the pixel-level segmentation result and the target detection box in parallel through the double-scale mask head, realizing end-to-end instance segmentation; S306, outputting the bud primordium mask.

4. The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 1, characterized in that, In S5, the contour fitting method used is the convex hull contour fitting.

5. The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 1, characterized in that, In S6, the morphological features of the bud primordium are the height-width ratio of the bud primordium, and the specific method for calculating is as follows: S601, extracting the convex hull vertex set; S602、Search for the two points with the farthest Euclidean distance in the convex hull vertex set (P max1 , max2 ) to form the main height axis L h : S603、Calculate L h vertical vector of where [Δx, Δy] = P max1 -P max2 ; S604, along V ⊥ directional projection convex hull vertex, determine the maximum width point pair (W start ,W end ): S605, calculate the bud base target morphological characteristics 6.The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 1, characterized in that, In S7, the growth phenotype features of the mushroom bud out-time period are the basic bud amount and the effective bud amount.

7. The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprout period according to claim 6, characterized in that, The total number of bud primordia N is divided by the number of the divided bud primordia to calculate the basic bud number. base is represented. 8.The intelligent recognition and calculation method for the growth phenotypic characteristics of the edible mushroom sprouting period according to claim 6, characterized in that, The calculation method of the effective bud amount is: counting the morphological characteristics of all bud primordia, and calculating the effective bud amount N according to the morphological characteristics R of the primordia aspect Screening the effective bud amount N meeting the requirements use ; The value of n is determined according to the growth standards of different types of mushrooms.

Images