Young fruit posture judgment method based on image recognition

Abstract

The invention discloses a young fruit posture judgment method based on image recognition, and belongs to the technical field of image recognition. According to the method, the corresponding young fruit analysis image is obtained by carrying out image acquisition on the specified young fruit area, illumination compensation and enhancement processing are carried out on the image, whether highlight fruit influence correction is carried out or not is synchronously judged to obtain the optimized young fruit analysis image, and if yes, young fruit contour extraction is carried out according to the optimized young fruit analysis image; and synchronously weakening the influence degree of the change of the young fruit features on contour extraction to obtain an adjusted young fruit analysis image, otherwise, marking the image as the adjusted young fruit analysis image, and then obtaining a fruit contour through the adjusted young fruit analysis image so as to carry out attitude fitting and output an attitude determination result. And finally, carrying out data extraction on the attitude determination result to carry out attitude determination verification, and carrying out attitude determination qualification feedback according to the verification result, thereby solving the problems of low attitude angle precision and difficulty in accurately reflecting the three-dimensional space attitude in the prior art.

Description

A method for determining the pose of young fruits based on image recognition Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method for determining the posture of young fruits based on image recognition. Background Technology

[0002] By introducing advanced technologies such as image recognition, deep learning, and computer vision, many limitations in traditional agricultural production models are broken, providing entirely new solutions for agricultural production. In current agricultural production, these intelligent technologies can not only significantly improve efficiency but also ensure crop quality, promoting efficient and sustainable agricultural development.

[0003] For example, Chinese invention patent CN117011843B discloses an image recognition and posture assessment method for automated dragon fruit harvesting, which includes: Step 1, forming an image dataset; Step 2, obtaining a dragon fruit image segmentation map; Step 3, converting the image segmentation map into a grayscale image and then into a binarized image; Step 4, obtaining the ecological posture vector and the fruit posture vector from the binarized image, and finally fitting the image to obtain the final posture of the fruit.

[0004] For example, Chinese invention patent CN115205852B discloses a method and device for thinning flowers and fruits based on image recognition, comprising: acquiring a first image of a fruit tree to be processed by an image acquisition device; determining the fruit tree type based on the first image; determining a flower and fruit thinning model matching the fruit tree type based on the fruit tree type; determining several target information and attribute values ​​of each target based on the flower and fruit thinning model; sending the first image, second images of each target, and attribute values ​​of each target to a preset expert planting model to determine the processing risk value of the target; and determining the processing target based on the attribute values ​​and corresponding processing risk values ​​of each target, so that the corresponding flower and fruit thinning management terminal of the image acquisition device can thin the flowers and fruits of the fruit tree to be processed according to the processing target.

[0005] The above technical solutions demonstrate the integration of modern agriculture and computer vision technology. Furthermore, automation technology reduces reliance on manual labor and improves the speed and accuracy of operations. At the same time, precise posture judgment, harvesting timing assessment, and fruit thinning operations can effectively improve fruit quality. In particular, the posture judgment of fruits can not only improve harvesting efficiency but also help farmers understand the growth status and distribution of crops in real time through data collection, thereby achieving refined planting management.

[0006] In the early stages of growth, young fruits are morphologically unstable and have significant surface reflections and occlusions, making it difficult for traditional posture determination methods to accurately identify their true posture. By combining computer vision with posture determination of young fruits, the posture angles of the fruits (such as tilt angle and principal axis direction) can be accurately obtained, providing data support for subsequent automated operations (such as harvesting and pruning). The current implementation process of this technology includes image acquisition, image preprocessing, young fruit detection and localization, posture feature extraction, posture determination, and result output and application.

[0007] Image acquisition utilizes high-definition industrial cameras or mobile terminals to capture images of young fruit trees. Multiple angles and lighting conditions can be set as needed to obtain complete visual information. Image preprocessing involves noise removal, illumination equalization, color space conversion, and image enhancement algorithms (such as histogram equalization and filtering) to improve the recognizability of target areas. Image segmentation or background removal is also performed to preserve the young fruit area. Young fruit detection and localization can automatically detect the location of young fruits using deep learning target detection algorithms, such as YOLO (You Only Look Once), Faster R-CNN (Faster Region-Convolutional Neural Network), and Mask R-CNN (Mask Region-Convolutional Neural Network). Networks (such as convolutional neural networks with masked regions) output the bounding box coordinates or segmentation mask for each young fruit, which is used for subsequent feature extraction. For overlapping or occluded young fruits, edge detection or contour analysis algorithms are used for separation. Pose feature extraction performs morphological analysis on the detected young fruit regions, extracting geometric features such as fruit outline, centroid, and major and minor axis directions. Key morphological parameters are calculated, such as the angle between the major axis and the horizontal direction (pose angle), the coordinates of the fruit's centroid, and the normal direction of the fruit surface (if using a depth map). For 3D data, an ellipsoidal model can be fitted using point clouds to obtain 3D pose parameters. Pose determination establishes pose classification rules or trains a classification model based on the extracted geometric and directional features. The results are visualized (e.g., angle labels are overlaid on the image) and can be uploaded to agricultural monitoring systems or control terminals for intelligent spraying angle adjustment, automatic harvesting robotic arm grasping position optimization, and statistical analysis of young fruit growth posture.

[0008] The above-mentioned technologies have at least the following technical problems: In the existing technology, under natural light conditions, the surface of fruits grown in greenhouses may be affected by strong light, shadows or reflected light spots, resulting in a decrease in image contrast and edge clarity. At the same time, uneven lighting can cause edge recognition algorithms to fail to reliably extract the boundaries of the fruit, especially for highly reflective fruits (such as apples and citrus). Moreover, young fruits have significant morphological variations at different growth stages and among different fruit tree varieties, and their surfaces often have uneven structures. Contour extraction algorithms (such as edge detection-based algorithms) may suffer from feature extraction errors due to irregularities or occlusions on the fruit surface, which in turn affects the calculation of posture angles, making it difficult to accurately identify the true shape of the fruit, leading to fitting errors, inaccurate posture angle calculations, and low posture angle accuracy, making it difficult to accurately reflect the posture in three-dimensional space. Summary of the Invention

[0009] To address the technical problem of low accuracy in posture angle determination of existing technologies, which leads to reduced reliability of young fruit posture assessment results, this invention provides a young fruit posture determination method based on image recognition. The technical solution is as follows: Images of a specified young fruit region are acquired to obtain corresponding young fruit analysis images. These images undergo illumination compensation and enhancement processing, and it is simultaneously determined whether to perform high-light fruit influence correction to obtain an optimized young fruit analysis image. If high-light fruit influence correction is performed, the young fruit contour is extracted based on the optimized young fruit analysis image, and the influence of changes in young fruit features on contour extraction is simultaneously weakened to obtain an adjusted young fruit analysis image. Otherwise, the young fruit analysis image is marked as an adjusted young fruit analysis image. Semantic segmentation and feature extraction are performed on the adjusted young fruit analysis image to obtain the corresponding fruit contour, which is then used for posture fitting to output the corresponding posture determination result. Data extraction is performed on the posture determination result to verify the posture determination, and feedback on the posture determination's pass / fail status is provided based on the verification result.

[0010] The beneficial effects of the technical solution provided by the embodiments of the present invention include at least the following: 1. The present invention provides a method for determining the posture of young fruit based on image recognition. By acquiring images of a specified young fruit area, the corresponding young fruit analysis image is obtained, which realizes the accurate acquisition of the target young fruit area and ensures that the image data corresponds one-to-one with the position of the target fruit. Then, the young fruit analysis image is subjected to illumination compensation and enhancement processing, and it is simultaneously determined whether to perform high-light fruit influence correction to obtain an optimized young fruit analysis image. This helps to improve the overall brightness balance of the image, improve the problem of local dark or bright areas caused by uneven illumination, and solve the problem of poor adaptability of traditional image enhancement algorithms to different lighting conditions.

[0011] 2. If highlight fruit influence correction is performed, the young fruit contour is extracted based on the optimized young fruit analysis image, and the influence of changes in young fruit features on contour extraction is weakened simultaneously to obtain an adjusted young fruit analysis image. This automatically weakens the interference of dynamic features (such as reflective points, shadow movement, and changes in epidermal texture) on contour extraction, which helps to avoid the problem of "discontinuity" or "virtual edges" in fruit boundaries caused by the susceptibility of traditional contour extraction algorithms to highlight interference. Otherwise, the young fruit analysis image is marked as the adjusted young fruit analysis image, ensuring the continuity of data flow processing and making up for the inconsistency of processing results under different image enhancement paths in traditional methods.

[0012] 3. By adjusting the young fruit analysis image for semantic segmentation and feature extraction, the corresponding fruit contour is obtained. This not only achieves accurate separation of the fruit region from the background, leaves, and branches, but also helps improve the integrity and accuracy of the fruit contour boundary. It reduces the probability of misjudging the contour when there is a large difference between the fruit and the ambient light. Then, the posture fitting is performed to output the corresponding posture judgment result, thereby improving the accuracy of posture fitting and thus improving the calculation accuracy of posture angles (such as principal axis direction angle and fruit stalk direction angle). Finally, the posture judgment result is extracted for posture judgment verification, and the posture judgment qualification feedback is given based on the verification result to ensure the reliability of the posture judgment result. This achieves automatic quality assessment of the posture judgment result, thereby ensuring that the posture estimation result meets the accuracy requirements, improving the reliability of the output result, and filling the gap of traditional systems that only output posture results and lack a result verification mechanism.

[0013] 4. This invention obtains corresponding correction degree factors by multiplying the correction degree values ​​of highly reflective fruits with the influence weights corresponding to the light compensation and enhancement treatment methods adopted. This achieves adaptive influence assessment of different light enhancement methods (such as Gamma correction, Retinex, etc.), helping to avoid over- or under-correction of reflectivity, improving the targeting and balance of enhancement treatment, and overcoming the shortcomings of traditional methods that use fixed parameters in light enhancement and cannot adaptively adjust according to reflectivity. Then, each correction degree factor is multiplied with the controllable parameters of each light compensation and enhancement treatment method to obtain the corresponding optimized controllable parameters, thus achieving... Automatic readjustment at the parameter level enables the illumination enhancement algorithm to adaptively adjust key parameters according to the degree of reflectivity, improving the versatility of the illumination compensation algorithm. This allows it to automatically obtain the optimal parameter combination under different lighting conditions. Finally, based on the optimized controllable parameters, the young fruit analysis image is re-processed with illumination compensation and enhancement to obtain an optimized young fruit analysis image. This allows the illumination compensation result to automatically tend towards the optimal under different reflectivity conditions, improving the overall brightness uniformity and detail contrast of the image. Consequently, it enhances the stability of subsequent contour extraction, semantic segmentation, and pose determination, solving the problem that the current technology lacks a secondary compensation mechanism for reflective areas, leading to a decrease in pose recognition accuracy.

[0014] 5. By performing semantic segmentation and feature extraction on the images of young fruits, the fruit contour is obtained, and then pose calculation is performed based on it. This achieves automatic separation of the fruit region from the background, improving the accuracy of subsequent feature extraction. Then, the pose calculation influence value is obtained by fusing the high reflectivity correction degree value and the young fruit difference analysis index, which more accurately quantifies the influence of each image on pose calculation. This solves the problem that traditional methods do not comprehensively consider multiple factors such as reflectivity and fruit differences, which leads to the accuracy of pose calculation being affected by uncontrollable factors in the image (such as highlights, dynamic changes, etc.). Finally, the pose calculation influence value is compared with the pose calculation threshold. This system automatically determines whether posture adjustment is needed, avoiding manual intervention. If the influence value of posture calculation exceeds the posture calculation threshold, corresponding posture fitting adjustment measures for young fruits are taken. This prevents the phenomenon that traditional methods do not automatically adjust posture fitting for young fruits with high reflectivity or large differences, which may lead to incorrect posture judgment results in these complex situations. This adjustment mechanism can reduce the risk of incorrect posture calculation and improve the accuracy of posture fitting in cases of severe reflectivity or unstable characteristics. If the influence value of posture calculation does not exceed the posture calculation threshold, posture calculation continues, avoiding unnecessary waste of computational resources and improving analysis efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 is a flowchart of an image recognition-based method for determining the posture of young fruit provided in an embodiment of this application; Figure 2 is a flowchart of a method for determining the influence of high reflectivity provided in an embodiment of this application; Figure 3 is a flowchart of a method for weakening the influence of changes in young fruit characteristics provided in an embodiment of this application; Figure 4 is a flowchart of a method for posture control provided in an embodiment of this application. Detailed Implementation

[0017] The following provides explanations for some of the terms used in this application. It should be noted that these explanations are for the convenience of those skilled in the art and do not constitute a limitation on the scope of protection claimed in this application.

[0018] The embodiments of this application involve at least one, including one or more; where "multiple" means two or more. Furthermore, it should be understood that in the description of this specification, terms such as "first," "second," and "third" are used only for descriptive purposes and should not be construed as indicating relative importance or order. For example, "first device" and "second device" do not represent the degree of importance of the two or their order, but are merely for descriptive distinction. In the embodiments of this application, "and / or" merely describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0019] The directional terms mentioned in the embodiments of this application, such as "up", "down", "left", "right", "inner", and "outer", are only for reference to the directions in the accompanying drawings. Therefore, the directional terms used are for better and clearer explanation and understanding of the embodiments of this application, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0020] References to "one embodiment," "in some examples," or "some embodiments" as described in the embodiments of this application mean that one or more embodiments of this specification include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in some examples," "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] As shown in Figure 1, Example 1 is a flowchart of an image recognition-based method for determining the posture of young fruit provided in this application. The method includes the following steps: Step 1: Illumination compensation and enhancement processing. Based on an industrial camera, images of a specified young fruit area are acquired to obtain corresponding young fruit analysis images. Then, illumination compensation and enhancement processing is performed on the young fruit analysis images, and it is simultaneously determined whether to perform high-light fruit influence correction to obtain optimized young fruit analysis images, so as to improve the reflection suppression effect of the young fruit analysis images.

[0023] The specific steps for obtaining optimized young fruit analysis images by correcting for the effects of high light on the fruit are as follows: Step 1, read the obtained young fruit analysis images and perform illumination compensation and enhancement processing. Illumination compensation and enhancement processing methods include one or more of the following: adaptive histogram equalization, illumination normalization, Gamma correction, reflection spot removal, and edge enhancement. The specific illumination compensation and enhancement processing methods and their corresponding quantities need to be determined based on the actual system scenario.

[0024] Step 2: Obtain high-reflectivity fruit data from the young fruit analysis image to determine the impact of high reflectivity and whether to perform high-reflectivity fruit correction. High-reflectivity fruit data includes brightness histogram skewness, brightness histogram kurtosis, brightness uniformity, and brightness entropy.

[0025] The specific method for obtaining the aforementioned high-reflectivity fruit data is as follows: The skewness of the image brightness values ​​is calculated using `scipy.stats.skew` to obtain the brightness histogram skewness; the kurtosis of the image brightness values ​​is calculated using `scipy.stats.kurtosis` to obtain the brightness histogram kurtosis; brightness uniformity is calculated by comparing the histogram differences between the original image and the equalized image. The specific calculation method is as follows: H(i) is the brightness histogram value of the original image, and E(i) is the brightness histogram value of the equalized image. i represents the gray level index, with a value range of i=0,1,2,...,L 1. L represents the brightness level, BE represents the brightness uniformity; brightness entropy is calculated using a formula based on a normalized histogram. The specific formula is as follows: p(i) is the probability of gray level index i appearing, i.e. the normalized value of the gray level histogram, L is the brightness level (usually 256, representing the brightness values ​​of 0-255 in the gray level image), and H(X) is the brightness entropy.

[0026] By jointly analyzing the illumination characteristics and high reflectivity features of young fruit images, adaptive optimization of illumination conditions and intelligent identification of high-light interference were achieved. First, various compensation and enhancement strategies, including adaptive histogram equalization, illumination normalization, gamma correction, reflected light spot removal, and edge enhancement, were employed to effectively improve image brightness distribution, enhance detail contrast, and improve the visual clarity and structural integrity of the fruit area. Subsequently, by calculating high-reflectivity fruit data such as skewness, kurtosis, brightness uniformity, and brightness entropy of the brightness histogram, potentially high-reflectivity areas in the image were quantitatively identified, enabling accurate identification and decision-making regarding the impact of high light. This method not only improves the targeting and intelligence of illumination correction but also effectively suppresses feature distortion caused by strong light or reflection, ensuring the stability and accuracy of subsequent fruit contour extraction and pose analysis, providing highly robust and high-quality visual input for young fruit image processing.

[0027] Figure 2 shows a flowchart of the high reflectivity impact judgment process provided in this application embodiment. The specific logic is as follows: the high reflectivity fruit data is compared with the corresponding judgment data. If the high reflectivity fruit data all meet the judgment conditions corresponding to the judgment data, then illumination compensation and enhancement continue; otherwise, the comprehensive impact quantification of the high reflectivity fruit data is performed to obtain the numerical value of the degree of influence of the high reflectivity fruit reflection phenomenon on the young fruit analysis image, which is recorded as the high reflectivity fruit influence degree coefficient, and then a judgment is made. The specific process is as follows: the high reflectivity fruit influence degree coefficient is projected into the high reflectivity degree correction projection sequence after training to obtain the corresponding high reflectivity fruit correction degree value; the high reflectivity fruit correction degree value is compared with the high reflectivity division interval to determine the controllable parameters for adjustment; through the above process, the illumination compensation and image enhancement processing process is made more accurate and dynamic, ensuring the quality of illumination compensation and image enhancement, and improving the robustness of image preprocessing.

[0028] Specifically, the process for judging the impact of high reflectivity is as follows: the data of highly reflective fruits are compared with the corresponding judgment data. The judgment data includes the skewness error range of the brightness histogram, the expected range of kurtosis, the reference value of brightness uniformity, and the reference value of brightness entropy. The judgment data are all obtained from the preset database, which is generally preset and stored by the preset staff based on historical data and experience rules.

[0029] It's worth noting that before designing an image recognition-based method for determining the posture of young fruits, technical professionals typically pre-build a pre-defined database to support various control strategies. This database integrates several key control parameters, including judgment data, high reflectivity correction projection sequences, high reflectivity segmentation intervals, influence weights, edge detection optimization tables, reference posture data, and posture calculation thresholds. All parameters are pre-set by technical professionals based on the selected analysis method and on-site hardware configuration. This pre-defined database provides the core data foundation for subsequent data uploading, storage optimization, and automated screening and judgment processes.

[0030] If the data of highly reflective fruits all meet the judgment conditions corresponding to the judgment data, then continue to perform illumination compensation and enhancement. The judgment conditions include that the skewness of the brightness histogram is within the skewness error range of the brightness histogram, the kurtosis of the brightness histogram is within the expected range of kurtosis, the brightness uniformity is greater than the brightness uniformity reference value (generally less than 1 but close to 1), and the brightness entropy is greater than the brightness entropy reference value.

[0031] Otherwise, the comprehensive impact quantification of the high reflectivity fruit data is used to obtain the numerical value of the influence of the reflectivity of the high reflectivity fruit on the analysis image of the young fruit. The data obtained by averaging the data after Z-Score standardization of the above high reflectivity fruit data is recorded as the high reflectivity fruit influence coefficient and is judged. The specific process is as follows: Based on the high reflectivity fruit influence coefficient, the high reflectivity correction projection sequence after training is projected to obtain the corresponding high reflectivity fruit correction value.

[0032] It should be added that by inputting the influence coefficient of highly reflective fruit into the high reflectivity correction projection sequence, the corresponding high reflectivity correction value can be obtained. This dataset is used to fit the mapping relationship between the influence coefficient of highly reflective fruit and the high reflectivity correction value. The construction method is as follows: in the initial data sequence constructed based on the gradient boosting regression algorithm, the influence coefficient of highly reflective fruit collected in the historical time period and the high reflectivity correction value set according to the empirical rules are selected as training samples. The model is trained based on the XGBoost framework with the least square error as the objective function, and finally the trained high reflectivity correction projection sequence is obtained.

[0033] The correction value of highly reflective fruit is compared with the set high reflectivity division intervals used to classify the impact of the reflectivity characteristics of highly reflective fruit on image analysis, in order to determine the controllable parameters for adjustment. Among them, the high reflectivity division intervals include mild division intervals, moderate division intervals, and severe division intervals, and the impact of the reflectivity characteristics of the corresponding highly reflective fruit on image analysis gradually increases. These are preset data set by the staff and are generally stored in a preset database.

[0034] It should be noted that the controllable parameters include one set of degree data and two sets of degree data. The effect of the control based on the one set of degree data is lower than the effect of the control based on the two sets of degree data. The specific process for determining the controllable parameters is as follows: If the correction degree value of the highly reflective fruit belongs to the mild division range, then the controllable parameters are adjusted based on the correction degree value of the highly reflective fruit. The controllable parameters are a set of degree data. The set of degree data includes image block size, multi-scale fusion ratio, brightness normalization coefficient, neighborhood window size, and filter kernel size.

[0035] If the correction degree value of highly reflective fruit belongs to the medium division range, then the controllable parameters are adjusted based on the correction degree value of highly reflective fruit. The controllable parameters are two sets of degree data; the two sets of degree data include cropping limit parameters, Gaussian filter scale, Gamma value, highlight threshold and edge detection threshold.

[0036] It needs to be explained that in image recognition-based young fruit pose determination, the clipping constraint parameter (Clip) The optimization effect of the limit is significantly higher than that of the image block size because it directly affects local contrast and edge gradient quality, while the core of pose recognition depends on the accuracy of edge and contour direction features. In illumination normalization, the adjustment of the Gaussian filter scale has a more significant effect on pose determination optimization because it directly affects the spatial scale of illumination compensation and edge gradient stability, while the multi-scale fusion ratio mainly improves the overall brightness consistency and has limited improvement on pose angle accuracy. Similarly, the adjustment of the Gamma value has a more significant effect on fruit pose determination optimization because it directly affects the contrast and edge gradient stability of the central brightness area of ​​the image, while the brightness normalization coefficient only adjusts the overall brightness and has limited substantial improvement on pose feature extraction. The adjustment of the high brightness threshold has a more significant effect on the optimization of most reflected light spots removal. It can quickly remove obvious reflected light spots, especially when the lighting conditions are highly contrasting, while the neighborhood window size plays a more moderate role in fine adjustments and complex backgrounds. The adjustment of the edge detection threshold has a more significant effect on pose determination optimization because it can accurately capture the key edge features of the fruit, while the adjustment of the filter kernel size has a more indirect effect on image noise removal and edge smoothing.

[0037] It should be added that the specific process of adjusting the controllable parameters based on the correction degree value of highly reflective fruit is as follows: First, the correction degree value of highly reflective fruit is multiplied by the influence weights corresponding to the illumination compensation and enhancement processing methods adopted, respectively, to obtain the corresponding correction degree factors; among them, the acquisition of influence weights also depends on the illumination compensation and enhancement processing methods adopted in the specific system scenario and the corresponding quantity, which are one or more of the following: adaptive histogram equalization influence weight, illumination normalization influence weight, Gamma correction influence weight, reflection spot removal influence weight, and edge enhancement influence weight.

[0038] Next, the corresponding optimized controllable parameters are obtained by multiplying each correction degree factor with the controllable parameters of each illumination compensation and enhancement processing method.

[0039] Finally, the young fruit analysis images were reprocessed with illumination compensation and enhancement based on optimized controllable parameters to obtain optimized young fruit analysis images.

[0040] For example, if the illumination compensation and enhancement processing method adopted is adaptive histogram equalization, illumination normalization, and Gamma correction, and the controllable parameters are two sets of degree data, then the corresponding controllable parameters are the clipping constraint parameter, the Gaussian filter scale, and the Gamma value. After multiplying each of them with the corresponding correction degree factor, the optimized controllable parameters can be obtained, namely the optimized clipping constraint parameter, the optimized Gaussian filter scale, and the optimized Gamma value.

[0041] Example B: If the illumination compensation and enhancement processing method adopted is reflected spot removal and edge enhancement, and the controllable parameters are a set of degree data, then the corresponding controllable parameters are the neighborhood window size and the filter kernel size. After multiplying each of them with the corresponding correction degree factor, the optimized controllable parameters can be obtained, namely the optimized neighborhood window size and the optimized filter kernel size.

[0042] If the correction value of highly reflective fruit falls within the severely defined range, the system will prompt the pre-set staff to re-acquire the young fruit analysis image.

[0043] By introducing a mechanism that combines the correction degree of highly reflective fruit with the influence weights of illumination compensation and enhancement processing methods, adaptive adjustment of the illumination optimization process is achieved. By weighting the correction degree factor with the controllable parameters of each processing method, optimized controllable parameters are obtained for different illumination conditions and fruit reflectivity characteristics, making the illumination compensation and image enhancement process more precise, dynamic, and personalized. This method can effectively suppress problems such as over-enhancement or distortion in reflective areas while ensuring image brightness balance, thereby improving the naturalness and detail fidelity of the image. Overall, the illumination correction process achieves intelligence and high robustness, providing a more stable and reliable visual foundation for subsequent fruit contour extraction and pose analysis.

[0044] Step 2: Weakening the impact of changes in young fruit features. If the influence of high-gloss fruit is corrected, the young fruit contour is extracted based on the optimized young fruit analysis image, and the impact of changes in young fruit features on contour extraction is weakened simultaneously to obtain an adjusted young fruit analysis image, so as to reduce the impact of dynamic changes in young fruit features on young fruit contour extraction. Otherwise, the young fruit analysis image is marked as the adjusted young fruit analysis image.

[0045] Figure 3 shows a flowchart of the process for weakening the influence of changes in young fruit characteristics provided in this application embodiment. The specific process is as follows: extract the young fruit analysis parameters of the corresponding type of young fruit from the preset database, perform comprehensive quantification based on the young fruit analysis parameters and edge detection parameters to obtain the corresponding young fruit difference analysis index, optimize the controllable parameters of edge detection according to the young fruit difference analysis index to obtain optimized controllable parameters, and perform edge detection on the optimized young fruit analysis image to obtain the adjusted young fruit analysis image. Through the above process, not only is the robustness and stability of edge detection improved, but the recognition error caused by differences in fruit shape is also reduced, ensuring the accuracy of the edge features of the adjusted young fruit analysis image.

[0046] The specific method for weakening the impact of the changes in young fruit characteristics on contour extraction is as follows: extract the young fruit analysis parameters of the corresponding type of young fruit from the preset database. The young fruit analysis parameters include the ratio of fruit stalk to fruit body, fruit peel color and texture features (grayscale statistics). The young fruit analysis parameters are average representative values ​​obtained based on historical data of the corresponding type of young fruit and the young fruit growth cycle of the currently specified young fruit area, as well as the growth pattern of the corresponding type of young fruit.

[0047] Based on the comprehensive quantification of young fruit analysis parameters and edge detection parameters, the corresponding young fruit difference analysis index is obtained. The young fruit difference analysis index is used to quantify the influence of the dynamic growth changes of young fruit on edge detection. The edge detection parameters include edge detectability (contrast) and posture calibration error.

[0048] Specifically, after Z-score standardization of the ratio of fruit stalk to fruit body, peel color (color entropy), and texture features (grayscale statistics), these factors are multiplied by their corresponding analysis weights and then summed to obtain the young fruit difference influence factor. The analysis weights include the ratio of fruit stalk to fruit body, peel color, and texture features, and the sum of these three is one, which is also read from a preset database. Next, the edge detection parameters are Z-score standardized, then mean-valued, and then multiplied by the young fruit difference influence factor to obtain the young fruit difference analysis index. The edge detectability (contrast) is calculated using the Sobel gradient mean, and the pose calibration error is calculated using Euclidean distance.

[0049] The controllable parameters for edge detection are optimized based on the young fruit difference analysis index to obtain optimized controllable parameters. Edge detection is then performed on the optimized young fruit analysis image to obtain the adjusted young fruit analysis image.

[0050] By introducing a comprehensive quantification mechanism combining database-based young fruit analysis parameters and edge detection features, dynamic adaptive optimization of the edge detection process is achieved. Feature parameters of different types of young fruits, including the ratio of fruit stalk to fruit body, peel color, and texture features, are extracted from a pre-set database, enabling accurate modeling of the morphological and appearance differences of target young fruits. Furthermore, these young fruit features are fused with edge detection parameters (such as edge detectability and pose calibration error) to form a quantified young fruit difference analysis index, used to characterize the impact of young fruit growth changes on edge detection accuracy. By adaptively adjusting the controllable parameters of edge detection based on this index, the parameters can be dynamically optimized according to changes in fruit morphology. This method effectively improves the robustness and stability of edge detection, reduces recognition errors caused by differences in fruit morphology, ensures the accuracy and continuity of edge features in the adjusted young fruit analysis image, and provides a reliable data foundation for subsequent pose calculation and morphological analysis.

[0051] The specific method for optimizing the controllable parameters of edge detection based on the young fruit difference analysis index is as follows: Query the edge detection optimization table set in advance in the preset database according to the young fruit difference analysis index to obtain the corresponding edge detection amplitude value. Based on the edge detection amplitude value, perform compensation calculations on the controllable parameters of edge detection to obtain the corresponding optimized controllable parameters. The optimized controllable parameters include optimizing the upper and lower limits of the Canny threshold and optimizing the Sobel kernel scale.

[0052] It should be added that the edge detection optimization table is used to characterize the correlation between the young fruit difference analysis index and the edge detection amplitude value. Its construction process is as follows: the young fruit difference analysis index collected in the historical time period and the edge detection amplitude value preset by professionals according to empirical rules are input into the initial data table constructed based on the logistic regression algorithm; then, the cross-entropy loss function is used as the optimization criterion, and the model is trained using the scikit-learn framework to finally generate the edge detection optimization table.

[0053] The third step is to perform posture fitting and output the corresponding posture determination results. By adjusting the young fruit analysis image, semantic segmentation and feature extraction are performed to obtain the corresponding fruit contour, so as to perform posture fitting and output the corresponding posture determination results, thereby improving the accuracy of posture fitting.

[0054] Specifically, the posture determination results include qualitative and quantitative results. Qualitative results include upright posture (stalk facing up), lateral posture (fruit tilted), inverted posture (stalk facing down), and horizontal posture (stalk approximately parallel to the horizontal plane). The system outputs a posture category label for each young fruit, which can be used for posture statistics, fruit tree growth monitoring, or harvesting strategy formulation. Quantitative results include tilt angle (angle between the fruit's main axis and the horizontal line), orientation angle (direction of rotation of the fruit in the plane), and if it is a 3D model, it outputs three-dimensional postures (Pitch, Yaw, Roll), which can be directly used for the grasping angle planning of agricultural robots or robotic arms.

[0055] Step 4: Attitude determination verification and pass / fail feedback. Data is extracted from the attitude determination results to verify the attitude determination, and the pass / fail feedback is given based on the verification results to ensure the reliability of the attitude determination results.

[0056] The above-mentioned data extraction for posture determination verification is carried out as follows: The posture determination results are extracted to obtain posture determination result data, which is then compared with the reference posture data of the corresponding type of young fruit extracted from the preset database. The posture determination result data includes the main axis direction angle, the fruit stalk direction angle, and the three-dimensional posture angle. The reference posture data includes the reference range of the main axis direction angle, the reference range of the fruit stalk direction angle, and the reference range of the three-dimensional posture angle. The preset data is set by the staff based on historical data, experience rules, and the growth pattern of young fruit within a historical time period, and is then entered into the preset database for storage.

[0057] If all the attitude determination results fall within the range of the corresponding reference attitude data, then the current attitude determination result is qualified, and the attitude determination of the young fruit analysis image continues.

[0058] If there is attitude determination result data that does not fall within the range corresponding to the reference data, it means that the current attitude determination result is unqualified, and then attitude determination qualification feedback is performed.

[0059] Among them, attitude determination qualification feedback means sending a prompt that the attitude determination of the current young fruit analysis image is unqualified to the terminal of the preset staff, and prompting the preset staff to perform image re-acquisition to obtain a secondary young fruit analysis image, and then performing young fruit attitude determination on the secondary young fruit analysis image again.

[0060] By introducing a comparison mechanism between the attitude determination result and the reference data, the automatic verification and feedback optimization of attitude analysis are realized. By extracting data from the attitude determination result, the spatial position and rotation information of the object are obtained, and then these results are compared with the reference attitude data extracted from the preset database. This comparison process can quantitatively judge whether the current attitude conforms to the preset reasonable range. If all the attitude determination result data are within the reasonable range of the corresponding reference attitude data, it means that the current attitude determination result is qualified, and the system will continue to perform subsequent attitude analysis work.

[0061] However, if some attitude determination result data do not conform to the range of the reference data, the system will determine it as unqualified, and through the attitude determination qualification feedback mechanism, automatically send a prompt message to the terminal of the staff, indicating that the attitude determination of the current young fruit analysis image is unqualified and requiring image re-acquisition. The staff can then perform re-acquisition according to the feedback to obtain a secondary young fruit analysis image and perform attitude determination on it again, so as to ensure the accuracy and reliability of attitude determination. Through this automated feedback mechanism, the need for manual intervention is reduced, and the accuracy and work efficiency of young fruit analysis are improved.

[0062] In this embodiment, aiming at the problems of uneven illumination, high-reflection fruit reflection, and unstable edge feature extraction during the acquisition process of young fruit images in a complex natural environment, a systematic image processing and attitude determination process is proposed, which has significant practical value and accuracy advantages; First, by collecting images of the specified young fruit area and performing illumination compensation and image enhancement processing, the problems of uneven brightness and shadow occlusion caused by natural light changes are effectively eliminated, the contrast and detail expression of the overall image are improved, and a stable visual foundation is laid for subsequent analysis. For the case of high-gloss fruits, the system automatically judges whether there is strong reflection interference and performs high-gloss influence correction, so that the surface features of the fruit skin can be truly restored, thus avoiding the errors caused by the high-gloss area to feature extraction and contour recognition.

[0063] Building upon image optimization, this application's method further integrates young fruit contour extraction and feature weakening techniques. By suppressing the interference of feature variations on contour detection, it improves the accuracy and completeness of fruit boundary detection, ensuring a high degree of consistency between the segmented region and the actual fruit morphology. Subsequently, the system utilizes semantic segmentation and feature extraction algorithms to obtain precise fruit contour information and performs pose fitting calculations based on this contour data, achieving high-precision estimation of the fruit's spatial pose (position and orientation). Through data extraction and verification of the pose results, this method can quantitatively evaluate the accuracy of pose determination, thereby achieving automatic feedback and optimization of the reliability of pose recognition.

[0064] Overall, this method achieves full automation and intelligence across the entire process, from image acquisition, illumination correction, contour extraction to pose determination and verification, which helps improve the stability and accuracy of young fruit image analysis. This process not only adapts to fruit image processing under complex lighting conditions but also helps reduce manual intervention and improve the consistency and reliability of pose recognition, providing important technical support and application value for fruit growth monitoring, automated grading, and intelligent agricultural visual analysis.

[0065] Example 2: Since existing traditional systems generally lack a mechanism to verify the obtained posture determination results, based on Example 1, an adjusted young fruit analysis image is obtained, followed by posture adjustment. The specific process is as follows: P1, semantic segmentation and feature extraction are performed on the adjusted young fruit analysis image to obtain the fruit outline, and then posture calculation is performed based on it. At the same time, the posture calculation influence value is obtained by fusing the high reflectivity fruit correction degree value and the young fruit difference analysis index. The fusion process means that the high reflectivity fruit correction degree value and the young fruit difference analysis index are multiplied by the corresponding high reflectivity influence weight and the young fruit difference influence weight, respectively, and then added. The sum of the high reflectivity influence weight and the young fruit difference influence weight is 1. This is preset by the staff based on the degree of influence of the two on the posture calculation and stored in the preset database.

[0066] It should be added that semantic segmentation methods include traditional methods such as threshold-based segmentation and edge detection (Canny, Sobel), or deep learning methods such as U-Net (U-shaped Convolutional Network), FCN (Fully Convolutional Network), and Mask R-CNN; methods for extracting color features include HSV (Hue, Saturation, Value), Lab (CIELAB Color Space), and color histograms; methods for extracting texture features include GLCM (Gray Level Co-occurrence Matrix), LBP (Local Binary Pattern), and Gabor filtering; methods for extracting shape features include Hu moments and fractal dimension; pose calculation can be achieved through various existing methods, such as the PnP (Perspective-n-Point Algorithm) algorithm based on 2D-3D matching or the DeepLabCut method in deep learning.

[0067] P2 compares the attitude calculation impact value with the attitude calculation threshold used to define the degree of influence of the current adjusted young fruit analysis image on the attitude calculation. The attitude calculation threshold is preset by a pre-set staff member.

[0068] P3. If the influence value of the pose calculation is higher than the pose calculation threshold, it means that the features of the young fruit in the current young fruit analysis image have a significant impact on the pose calculation. In this case, corresponding young fruit pose fitting and control measures will be taken.

[0069] P4. If the influence value of pose calculation is not higher than the pose calculation threshold, it means that the young fruit features of the current young fruit analysis image have little influence on pose calculation, and pose calculation continues.

[0070] Figure 4 shows a schematic diagram of the posture control process provided in this application embodiment. The specific logic is as follows: semantic segmentation and feature extraction are performed on the young fruit analysis image to obtain the fruit outline, and then posture calculation is performed based on it. The posture calculation influence value is obtained by fusing the high reflectivity fruit correction degree value and the young fruit difference analysis index. The posture calculation influence value is compared with the posture calculation threshold used to define the influence of the current young fruit analysis image on the posture calculation: if the posture calculation influence value is higher than the posture calculation threshold, the corresponding young fruit posture fitting and control measures are taken; if the posture calculation influence value is not higher than the posture calculation threshold, the posture calculation continues. The above process not only ensures the high accuracy and stability of posture determination, but also improves the dynamic control capability of posture calculation.

[0071] Secondly, the specific content of the posture fitting and regulation measures for young fruits is as follows: P31, based on the deviation between the posture calculation influence value and the posture calculation threshold, the corresponding posture calculation regulation ratio is obtained; it should be noted that the difference between the posture calculation influence value and the posture calculation threshold is recorded as the posture calculation difference quantity. The posture calculation difference quantity is input into the posture calculation regulation queue to obtain the corresponding posture calculation regulation ratio. This queue is used to fit the mapping relationship between the posture calculation difference quantity and the posture calculation regulation ratio. Its construction method is as follows: in the initial data queue built based on the linear regression algorithm, the posture calculation difference quantity collected in the historical time period and the posture calculation regulation ratio set according to the empirical rules are selected as training samples. The least squares method criterion is used and the model is trained based on the statsmodels framework to finally obtain the trained posture calculation regulation queue.

[0072] P32 shows that the attitude calculation parameters are adjusted by the attitude calculation control ratio to obtain the optimized attitude calculation parameters. This involves multiplying the attitude calculation control ratio with the attitude calculation parameters, and then performing young fruit attitude fitting based on the optimized attitude calculation parameters, and outputting the corresponding attitude judgment results. The attitude calculation control ratio includes the ellipse fitting error tolerance control ratio, the Kalman filter noise covariance control ratio, the time window length control ratio, and the contour missing compensation ratio control ratio.

[0073] It should be noted that common pose fitting methods are as follows: (1) Pose fitting based on 2D-3D points: PnP algorithm: using 3D-2D point pairs to calculate rotation matrix and translation vector; EPnP (Efficient Perspective-n-Point Algorithm) and UPnP (Unified Perspective-n-Point Algorithm): optimized versions of PnP algorithm, improving accuracy.

[0074] (2) Pose fitting based on deep learning: PoseNet (Pose Estimation Network): Based on convolutional neural network, it directly predicts 6-DOF pose; DensePose (Dense Human Pose Estimation): It uses deep learning to densely predict the 3D position of each pixel, which is suitable for complex pose estimation.

[0075] (3) Geometric constraint-based attitude fitting: ICP (Iterative Closest Point) algorithm: Aligns point clouds by iteratively minimizing the distance between points, suitable for 3D point cloud data.

[0076] (4) Parallax-based pose fitting: Binocular vision: Depth is calculated through the parallax map to infer 3D pose.

[0077] In this embodiment, the accuracy and adaptability of pose calculation are further optimized by introducing a fusion process based on the correction degree of highly reflective fruit and the young fruit difference analysis index. First, the adjusted young fruit analysis image undergoes semantic segmentation and feature extraction to accurately obtain the fruit's contour information, providing a stable foundation for subsequent pose calculation. Next, by combining the correction degree of highly reflective fruit and the young fruit difference analysis index, the pose calculation influence value is calculated, comprehensively quantifying the potential impact of young fruit image features on the pose fitting results, ensuring the accuracy and reliability of the calculation results.

[0078] Based on this, the system compares the influence value calculated by attitude with a preset attitude calculation threshold to determine the degree of influence of the current image. If the influence value exceeds the threshold, the system automatically initiates attitude fitting and adjustment measures to further adjust the attitude calculation model, ensuring high accuracy and stability of attitude determination. If the influence value does not exceed the threshold, the system continues with routine attitude calculation, avoiding unnecessary overprocessing. Overall, this method effectively improves the dynamic adjustment capability of attitude calculation through intelligent influence value evaluation and threshold judgment, ensuring that it maintains high determination accuracy even under complex lighting and reflection conditions, providing strong technical support for agricultural automation and visual monitoring.

[0079] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)). Where there is no conflict, the solutions in the above embodiments can be combined.

[0080] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0084] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope and intent of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A method for determining the posture of young fruits based on image recognition, characterized in that, Includes the following steps: Image acquisition is performed on a designated young fruit area to obtain corresponding young fruit analysis images. Illumination compensation and enhancement processing are applied to these images, and it is simultaneously determined whether to perform highlight fruit influence correction to obtain an optimized young fruit analysis image. If highlight fruit influence correction is performed, the young fruit contour is extracted based on the optimized young fruit analysis image, and the influence of changes in young fruit features on contour extraction is simultaneously weakened to obtain an adjusted young fruit analysis image. Otherwise, the young fruit analysis image is marked as an adjusted young fruit analysis image. Semantic segmentation and feature extraction are performed on the adjusted young fruit analysis image to obtain the corresponding fruit contour, which is then used for pose fitting to output the corresponding pose determination result. Data extraction is performed on the pose determination result to verify the pose determination, and the pose determination pass / fail feedback is provided based on the verification result.

2. The method for determining the posture of young fruit based on image recognition as described in claim 1, characterized in that: The specific steps for obtaining optimized young fruit analysis images through high-light fruit influence correction are as follows: Step 1, read the obtained young fruit analysis image and perform illumination compensation and enhancement processing. The illumination compensation and enhancement processing methods include one or more of the following: adaptive histogram equalization, illumination normalization, Gamma correction, reflection spot removal, and edge enhancement; Step 2, obtain high-reflectivity fruit data from the young fruit analysis image to determine the influence of high reflectivity, thereby deciding whether to perform high-light fruit influence correction. The high-reflectivity fruit data includes brightness histogram skewness, brightness histogram kurtosis, brightness uniformity, and brightness entropy.

3. The method for determining the posture of young fruit based on image recognition as described in claim 2, characterized in that: The specific process for judging the impact of high reflectivity is as follows: the data of highly reflective fruits are compared with the corresponding judgment data, which includes the skewness error range of the brightness histogram, the expected range of kurtosis, the reference value of brightness uniformity, and the reference value of brightness entropy. If the data of highly reflective fruits all meet the judgment conditions corresponding to the judgment data, then continue to perform light compensation and enhancement; Otherwise, a comprehensive impact quantification of the high-reflectivity fruit data is performed to obtain a numerical value representing the degree of influence of the high-reflectivity fruit's reflectivity on the image analysis of young fruit. This value is denoted as the high-reflectivity fruit influence coefficient and is then determined. The specific process is as follows: Based on the high-reflectivity fruit influence coefficient, the high-reflectivity correction projection sequence at the end of training is projected to obtain the corresponding high-reflectivity fruit correction value. The high-reflectivity fruit correction value is compared with the set high-reflectivity division intervals used to classify the degree of influence of the reflectivity characteristics of high-reflectivity fruits on image analysis to determine the controllable parameters. The high-reflectivity division intervals include mild, moderate, and severe division intervals, with the degree of influence of the reflectivity characteristics of high-reflectivity fruits on image analysis gradually increasing. The controllable parameters include one set of degree data and two sets of degree data, where the effect of the control based on the one set of degree data is lower than the effect of the control based on the two sets of degree data.

4. The method for determining the posture of young fruit based on image recognition as described in claim 3, characterized in that: The specific process for determining the controllable parameters is as follows: If the correction degree value of highly reflective fruit belongs to the mild division range, the controllable parameters are adjusted based on the correction degree value of highly reflective fruit, and the controllable parameters are a set of degree data. If the high-reflective fruit correction degree value belongs to the moderate division interval, the controllable parameters are regulated based on the high-reflective fruit correction degree value, and the two groups of degree data are the controllable parameters to be regulated; If the high-reflective fruit correction degree value belongs to the severe division interval, it is prompted that the preset staff conduct re-sampling of the young fruit analysis image.

5. The method for determining the posture of young fruit based on image recognition as described in claim 4, characterized in that: The specific process of regulating the controllable parameters based on the high-reflective fruit correction degree value is as follows: Multiply the high-reflective fruit correction degree value by the corresponding influence weights of the adopted light compensation and enhancement processing methods respectively to obtain the corresponding correction degree factors; Multiply each correction degree factor by the controllable parameters of each light compensation and enhancement processing method to obtain the corresponding optimized controllable parameters; Based on the optimized controllable parameters, re-perform light compensation and enhancement processing on the young fruit analysis image to obtain an optimized young fruit analysis image.

6. The method for determining the posture of young fruit based on image recognition as described in claim 1, characterized in that: The specific method for synchronously weakening the influence degree of the change of young fruit characteristics on contour extraction is as follows: Extract the young fruit analysis parameters of the corresponding type of young fruit from the preset database, and the young fruit analysis parameters include the fruit stalk to fruit body ratio, fruit skin color, and texture characteristics; Based on the comprehensive quantification of the young fruit analysis parameters and the edge detection parameters, the corresponding young fruit difference analysis index is obtained. The young fruit difference analysis index is used to quantify the influence degree of the dynamic growth change of the young fruit on edge detection. The edge detection parameters include edge detectability and pose calibration error; Optimize the controllable parameters of edge detection according to the young fruit difference analysis index to obtain optimized controllable parameters, and perform edge detection on the optimized young fruit analysis image to obtain an adjusted young fruit analysis image.

7. The method for determining the posture of young fruit based on image recognition as described in claim 6, characterized in that: The specific method for optimizing the controllable parameters of edge detection according to the young fruit difference analysis index is as follows: Query according to the young fruit difference analysis index in the set edge detection optimization table to obtain the corresponding edge detection amplitude value, and perform compensation operations on the controllable parameters of edge detection respectively based on the edge detection amplitude value to obtain the corresponding optimized controllable parameters. The optimized controllable parameters include the upper and lower limits of the optimized Canny threshold and the optimized Sobel kernel scale.

8. The method for determining the posture of young fruit based on image recognition as described in claim 1, characterized in that: The specific process of extracting data from the pose determination result for pose determination verification is as follows: Extract data from the pose determination result to obtain pose determination result data, and then compare it with the reference pose data extracted from the preset database respectively; If the pose determination result data all belong to the range corresponding to the corresponding reference pose data, it means that the current pose determination result is qualified, and continue to perform pose determination on the young fruit analysis image; If there is pose determination result data that does not belong to the range corresponding to the corresponding reference data, it means that the current pose determination result is unqualified, and then perform pose determination qualification feedback; The pose determination qualification feedback means sending a prompt that the pose determination of the current young fruit analysis image is unqualified to the terminal of the preset staff, and prompting the preset staff to perform image re-sampling to obtain a secondary young fruit analysis image, and then perform young fruit pose determination on the secondary young fruit analysis image again.

9. A method for determining the posture of young fruit based on image recognition as described in any one of claims 1-8, characterized in that: The process of obtaining the adjusted young fruit analysis image includes posture control, which is as follows: semantic segmentation and feature extraction are performed on the adjusted young fruit analysis image to obtain the fruit outline, and then posture calculation is performed based on it. The posture calculation influence value is obtained by fusing the high reflectivity fruit correction degree value and the young fruit difference analysis index. The posture calculation influence value is compared with the posture calculation threshold used to define the influence of the current adjusted young fruit analysis image on the posture calculation: if the posture calculation influence value is higher than the posture calculation threshold, corresponding young fruit posture fitting and control measures are taken; if the posture calculation influence value is not higher than the posture calculation threshold, posture calculation continues.

10. The method for determining the posture of young fruit based on image recognition as described in claim 9, characterized in that: The specific details of the young fruit posture fitting and control measures are as follows: A mapping is performed between the deviation between the posture calculation influence value and the posture calculation threshold to obtain the corresponding posture calculation control ratio; the posture calculation parameters are adjusted using the posture calculation control ratio to obtain optimized posture calculation parameters; young fruit posture fitting continues based on the optimized posture calculation parameters, and the corresponding posture judgment results are output; the posture judgment results include qualitative and quantitative judgment results; the qualitative judgment results are used for posture statistics, fruit tree growth monitoring, or harvesting strategy formulation; the quantitative judgment results are directly used for the grasping angle planning of agricultural robots or robotic arms. The attitude calculation control ratios include the ellipse fitting error tolerance control ratio, the Kalman filter noise covariance control ratio, the time window length control ratio, and the contour loss compensation ratio control ratio.

Images