Camellia oleifera multi-objective intelligent detection method and system with dynamic light self-adaptive adjustment

Abstract

The application discloses a kind of dynamic light self-adapting regulation's camellia multi-objective intelligent detection method and system, by the region segmentation pretreatment of to-be-detected camellia image, generate multi-region segmentation camellia image;Multi-region segmentation camellia image is input into camellia fruit multi-objective recognition immune algorithm, by the light intensity analysis of multi-region segmentation camellia that has been trained good preference multi-objective module;Among them, preference multi-objective module is used to the form color value of camellia fruit under different illumination as preference data carries out target identification;According to the light intensity analysis result that has been trained good preference multi-objective module exports, obtain the recognition result of camellia fruit in to-be-detected camellia image.The application guarantees efficiency and improves speed on dynamic light recognition, provides effective solution for camellia fruit recognition under natural environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and particularly discloses a dynamic light self-adaptive adjustment camellia multi-target intelligent detection method and system. BACKGROUND

[0002] At present, in the field of camellia production and processing automation, target visual recognition, especially visual recognition under natural light, is a very important link. Only by recognizing target camellia fruits to the greatest extent in a natural environment can it be conducive to subsequent unmanned picking by picking machines, thereby saving production and processing costs.

[0003] In recent years, many scholars at home and abroad have conducted a large amount of research on intelligent algorithms, which have been widely applied in the field of agricultural and forestry target recognition. In the field of intelligent target recognition of camellia, a large amount of research has also been conducted in recent years. Chen Fengjun et al. proposed an improved YOLOv7 algorithm to detect the maturity of camellia fruits, introduced a cross-attention mechanism to optimize the backbone network, and improved the intersection over union. After training, the average detection time was 0.77s, proving the usability of the algorithm. Xie Weijun et al. used AlexNet as the backbone network to solve the problem of existing camellia color sorter recognition of broken seeds, and optimized the number of layers of the subsequent network, finally completed the integrity color selection recognition of camellia broken seeds, and the recognition rate was greatly improved compared with the traditional method. Zhou Hongping et al. used YOLOv8n recognition algorithm to classify and recognize camellia fruits. This method used coco dataset as the training model, combined 4 factors such as learning method, training data volume, learning rate and training number of rounds, and conducted 52 groups of YOLOv8n ablation experiments, finally achieved better results; Meng Zhichao et al. proposed an improved VGG16 network to solve the problem of chaotic camellia seedling varieties on the market, and completed the training and testing on the camellia leaf dataset; Wang Jinping et al. used SCConv as the backbone network YOLOv8n algorithm to identify the occlusion phenomenon of camellia, which led to low recognition rate; Duan Yufei et al. conducted research on the visual sorting of camellia shelling machine fruit shell and seeds, used VGG16 as the backbone network and optimized it, which relieved the gradient explosion problem of image detection, and thus improved the system robustness.

[0004] There are many scholars in the field of agriculture and forestry at home and abroad who have done a lot of research on light and shelter identification affected by light. Liu et al. used the improved Faster R-CNN method for the identification of multi-cluster green persimmons under different light conditions in natural environment. YANG et al. used a double CNN combined method to enhance low-light images to realize image reconstruction. Du Jinzhi et al. used a low-light enhancement method to solve the problem of light, which was to use an improved GAN to enhance the image of kiwi flower under low-light conditions, thereby realizing the brightness enhancement of kiwi flower image under low-light conditions, and effectively suppressing the occurrence of noise. Li Xu et al. added a fusion efficient channel CA (coordinate attention) attention mechanism in the YOLOX algorithm module to improve the ability of the algorithm to identify the main features of pepper fruits under different light conditions. Zhang Zhonglili et al. used the R-G method to enhance the red area on the surface of the tomato, and then used the Otsu method to extract and process the surface coloring area, and further used the local brightness equalization method to repair and solve the high brightness reflection of the tomato surface caused by light changes. The average accuracy of the colored area on the surface of the tomato reached 92.96%. Wang Chuanyu et al. proposed a dynamic light removal method based on multi-exposure image fusion. This method maps different light radiation intensities to the RGB space after calculating the light radiation intensity of the corn canopy. The brightness of the hemispherical image is evenly distributed after mapping, thereby eliminating the influence of light changes on the image. Bao Wenxia et al. used the Retinex algorithm to enhance the image of wheat taken by a drone. And add coordinate attention mechanism in the main network of YOLOv5 recognition to refine the model features. Through the test, compared with the basic recognition algorithm, the average accuracy is improved by 4.1%.

[0005] Although the above methods have achieved certain results in the detection of agricultural and forestry targets and light, they do not optimize the light in the dynamic changes of the external environment in nature. Camellia oleifera picking time is generally between September and November every year. At present, camellia oleifera has basically realized automation in the field of picking and processing, but there are still many problems in the process of automatic recognition. The main problem is that in the natural environment, the light conditions of camellia oleifera picking base are complex, and the light has a great influence on the recognition rate. At the same time, the recognition rate of different camellia oleifera tree areas is not the same, which leads to the situation that the recognition rate is not the same in the morning, noon and evening after the same recognition algorithm is debugged, and the recognition rate is high in a certain area of camellia oleifera tree, but the recognition rate is low in another area. The dynamic change of light intensity in the natural environment brings great difficulty to intelligent picking and recognition. SUMMARY

[0006] The present application provides a kind of dynamic light self-adapting regulation's camellia oleifera multi-target intelligent detection method and system, to solve at least one of the defects existing in the prior art.

[0007] One aspect of the present application relates to a dynamic light adaptive adjustment camellia multi-objective intelligent detection method, comprising the following steps:

[0008] The image to be detected is subjected to region segmentation preprocessing to generate a multi-region segmented camellia image;

[0009] The multi-region segmented camellia image is input into the camellia fruit multi-objective recognition immune algorithm, and the light intensity of the multi-region segmented camellia image is analyzed by the trained preference multi-objective module; wherein the preference multi-objective module is used for target recognition of the morphological color values of the camellia fruit under different light conditions as preference data;

[0010] According to the light intensity analysis result output by the trained preference multi-objective module, the recognition result of the camellia fruit in the image to be detected is obtained.

[0011] Further, before the step of inputting the multi-region segmented camellia image into the camellia fruit multi-objective recognition immune algorithm and analyzing the light intensity of the multi-region segmented camellia image by the trained preference multi-objective module, the step includes:

[0012] Dynamic light intensity data of camellia fruit under different light conditions is collected, including light intensity under different weather and different time periods;

[0013] The morphological color values corresponding to the dynamic light intensity data are extracted, and the morphological color values are input into the trained preference multi-objective module as preference data for affinity clustering analysis training; the morphological color values include RGB color values and morphological values; the preference data includes preference recognition data and preference levels.

[0014] Further, the step of generating a multi-region segmented camellia image by region segmentation preprocessing of the image to be detected includes:

[0015] The image to be detected is subjected to region segmentation preprocessing, and the image to be detected is segmented into several regions, each region corresponding to a matched preference level;

[0016] It is judged whether the size of the remaining image after segmentation of the image to be detected meets the preset segmentation threshold, if the size of the remaining image after segmentation of the image to be detected is less than the preset segmentation threshold, the remaining image after segmentation of the image to be detected is merged into the adjacent region.

[0017] Further, in the step of inputting the multi-region segmented camellia image into the camellia fruit multi-objective recognition immune algorithm and analyzing the light intensity of the multi-region segmented camellia image by the trained preference multi-objective module, the affinity clustering results of the camellia fruit targets in each region are integrated according to the immune affinity formula, the camellia fruit targets are recognized from the data and the coordinate values are returned, and the immune affinity formula is:

[0018]

[0019] wherein, is an affinity value, is a Gaussian function symbol, is a Gaussian function parameter, set as the input camellia fruit R component, G component, B component, elongation, circularity and roundness 6 preferred feature parameters; is a to-be-detected camellia image.

[0020] Further, in the step of obtaining the recognition result of the camellia fruit in the to-be-detected camellia image according to the light intensity analysis result output by the trained preference multi-objective module, based on the light intensity analysis result, the new camellia population is obtained by performing selection, crossover and mutation operations on the camellia antibody population, the cell affinity is calculated in the new camellia population, and the cell affinity is:

[0021]

[0022] wherein, is a cell a new cell formed after mutation, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is an exponential decay variable in the regulation function, is a cell adaptive value after standardization processing, is a self-regulating variation coefficient.

[0023] Another aspect of the present application relates to a dynamic light adaptive adjustment camellia multi-objective intelligent detection system, comprising:

[0024] A generation module is configured to perform region segmentation preprocessing on the to-be-detected camellia image to generate a multi-region segmented camellia image.

[0025] An analysis module is configured to input the multi-region segmented camellia image into a camellia fruit multi-objective recognition immune algorithm, and perform light intensity analysis on the multi-region segmented camellia image by a trained preference multi-objective module; the preference multi-objective module is configured to perform target recognition on the morphological color values of the camellia fruit under different light as preference data.

[0026] An acquisition module is configured to obtain the recognition result of the camellia fruit in the to-be-detected camellia image according to the light intensity analysis result output by the trained preference multi-objective module.

[0027] Further, the dynamic light adaptive adjustment camellia multi-objective intelligent detection system further comprises:

[0028] The collection module is configured to collect dynamic light intensity data of the tea fruit under different illuminations, wherein the different illuminations include different weather and different time periods of illumination intensity;

[0029] The extraction module is configured to extract morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into the preference multi-objective module to be trained for affinity clustering analysis training, wherein the morphological color values include RGB color values and morphological values, and the preference data includes preference identification data and preference levels.

[0030] Further, the generation module includes:

[0031] The segmentation unit is configured to perform region segmentation preprocessing on the to-be-detected tea image, and divide the to-be-detected tea image into a plurality of regions, each region corresponding to a matched preference level.

[0032] The judgment unit is configured to judge whether the size of the remaining image after segmentation of the to-be-detected tea image meets a preset segmentation threshold, and if the size of the remaining image after segmentation of the to-be-detected tea image is less than the preset segmentation threshold, the remaining image after segmentation of the to-be-detected tea image is merged into a neighboring region.

[0033] Further, in the analysis module, the affinity clustering results of the tea fruit targets in each region are comprehensively analyzed according to an immune affinity formula, the tea fruit targets are identified from the data, and coordinate values are returned, and the immune affinity formula is:

[0034]

[0035] wherein, is an affinity value, is a Gaussian function symbol, is a Gaussian function parameter, is set as six preference characteristic parameters of the input tea fruit R component, G component, B component, elongation rate, circularity and roundness; is the to-be-detected tea image.

[0036] Further, in the acquisition module, based on the light intensity analysis result, the tea antibody population is selected, crossed and mutated to obtain a new tea population, and cell affinity is calculated in the new tea population, and the cell affinity is:

[0037]

[0038] wherein, is a cell a new cell formed after mutation, is an original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is an exponential decay variable in a regulation function, is a cell adaptive value after standardization processing, is a self-adjusting coefficient of variation.

[0039] The beneficial effects achieved by the present application are:

[0040] The present application provides a kind of dynamic light adaptive regulation camellia oleifera multi-objective intelligent detection method and system, the image of camellia oleifera to be detected is carried out region segmentation preprocessing, generates multi-region segmentation camellia oleifera image;Multi-region segmentation camellia oleifera image is input into camellia oleifera fruit multi-objective recognition immune algorithm, and the light intensity analysis of multi-region segmentation camellia oleifera is carried out by the trained preference multi-objective module;Wherein, preference multi-objective module is used to carry out target identification to the form color numerical value of camellia oleifera fruit under different illumination as preference data;According to the light intensity analysis result output by the trained preference multi-objective module, the recognition result of camellia oleifera fruit in the image to be detected is obtained.The dynamic light adaptive regulation camellia oleifera multi-objective intelligent detection method and system provided by the present application achieve the following beneficial effects:

[0041] 1) the recognition rate is improved by multi-objective preference data optimization, and the 6 characteristics of the form and color of camellia oleifera fruit under different light intensities corresponding to the preference data, solve the problem of dynamic change of target recognition rate under uneven illumination intensity.

[0042] 2) the efficiency of immune algorithm is improved by the improvement and optimization of preference level, image partition and other immune algorithms, and the efficient algorithm structure of immune algorithm is used to make it suitable for natural picking environment of camellia oleifera with high real-time rate requirement and low-cost software and hardware system.

[0043] 3) compared with 5 kinds of algorithms of traditional immune algorithm (Classic-AIS), traditional genetic algorithm (GA), Faster R-CNN, BP neural network (BP-Network) and YOLOv8n, the recognition rate of the algorithm proposed in this paper under natural light reaches 93.8%, and the recognition time is 160ms, which proves the availability of the algorithm under dynamic light.

[0044] The experiment and test of the research model are based on more than 4500 test images under different illumination conditions, and the research is mainly aimed at the identification of color, shape feature preference characteristics of camellia oleifera fruit. In dynamic light recognition, both efficiency and speed are guaranteed, which provides an effective solution for camellia oleifera fruit recognition in natural environment. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a flowchart of an embodiment of the present application, a dynamic light adaptive regulation camellia oleifera multi-objective intelligent detection method;

[0046] Figure 2 A solution identification diagram for camellia target recognition in a dynamic light adaptive adjustment camellia multi-target intelligent detection method of the present application. DETAILED DESCRIPTION

[0047] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings of the specification and specific embodiments.

[0048] As shown in Figure 1 The first embodiment of the present application proposes a dynamic light adaptive adjustment camellia multi-target intelligent detection method, which includes the following steps:

[0049] Step S100, the camellia image to be detected is subjected to region segmentation preprocessing to generate a multi-region segmented camellia image.

[0050] The camellia image to be detected is subjected to region segmentation, and pixels are selected for region segmentation detection, and the number of regions can be adjusted according to the focal length of the camera and the distance of the collection.

[0051] Step S200, the multi-region segmented camellia image is input into a camellia fruit multi-target recognition immune algorithm, and the light intensity of the multi-region segmented camellia image is analyzed by a trained preference multi-target module; wherein the preference multi-target module is used for target recognition of the shape and color values of the camellia fruit under different lightings as preference data.

[0052] The generated multi-region segmented camellia image is input into the camellia fruit multi-target recognition immune algorithm for affinity clustering analysis, and the preference multi-target module in the camellia fruit multi-target recognition immune algorithm is used to fuse the typical preference recognition data of the shape and color corresponding to the dynamic light intensity data detected by the light meter, the preference level and the region segmentation module, and to recognize the camellia fruit under different light intensities.

[0053] Step S300, according to the light intensity analysis result output by the trained preference multi-target module, the recognition result of the camellia fruit in the camellia image to be detected is obtained.

[0054] After each segmented region inputs the preference data, the affinity clustering results of the camellia fruit targets of each region are integrated, the camellia fruit target is identified, and the coordinate value is returned.

[0055] Further, the dynamic light adaptive adjustment camellia multi-target intelligent detection method provided by the present embodiment includes the following steps before step S200:

[0056] Step S200A, dynamic light intensity data of camellia fruits under different lightings is collected, and the different lightings include different weather and different time periods of light intensity.

[0057] The collection device is SONY ZV-E10L micro single digital camera, and the camera is equipped with a professional lens Nikkor 18-50 to obtain better image quality. The collected oil tea variety is Xianglin-1 oil tea.

[0058] The sampling time is sunny and cloudy days, and the oil tea pictures are collected from 9 am to 21 pm under natural light. According to the light conditions, the collection is carried out every 1 hour or so, a total of 12 collection time points, and an average of 570 pictures are collected for each light intensity to facilitate subsequent comprehensive analysis. The illumination range of the picture is from 50 lux to 100,000 lux, and the oil tea fruit to be collected target in natural environment.

[0059] From the analysis of the collected oil tea fruit target image, it can be seen that the brightness of the oil tea fruit pictures grown in natural environment is not only greatly affected by the light angle, but also the mutual shielding between tree branches, leaves and fruits will affect the picture quality. This complex natural environment brings great difficulty to the practicalization of forestry picking visual system, and is also a problem that must be solved for the realization of automatic oil tea visual system.

[0060] The luxmeter is also called the luxmeter, which is an instrument for measuring light intensity, measuring the degree of illumination of the object surface, that is, the ratio of light flux and illuminated area. The light intensity measuring device selected in this test is Delex LSK-2304 high-precision luxmeter, with an illumination range of 0-200,000 lux, a resolution of 1 lux, an operating temperature environment of 0-40 degrees Celsius, a high-precision photosensitive probe, a temperature detection function, an illumination measurement accuracy of plus or minus 4%, and a history of illumination data storage. Its data index can be applied to the dynamic natural light environment of oil tea, and the measuring time needs to be fixed to the direction of the image to be collected to obtain stable light intensity data.

[0061] Considering the influence of weather on light, the measurement time is divided into sunny and cloudy days, and the light intensity is measured from 9 am to 21 pm in two weather conditions. Since the light intensity is dynamic, the illumination is measured for two minutes each time, and the average value is taken every 10 seconds.

[0062] The most influential factor for the target detection of Camellia oleifera fruits grown in natural environment is light intensity. The recognition rate of pictures is not high under normal sunny noon or cloudy early morning or late afternoon. The data indicators of cloudy noon are similar to those of sunny early morning or late afternoon, and the recognition degrees are higher. After analyzing more than 4500 pictures collected under different light intensities, the pictures are divided into 8 levels according to the illumination value, i.e. more than 40,000 lux, 10,000-40,000 lux, 8,000-10,000 lux, 4,000-8,000 lux, 1,000-4,000 lux, 300-1,000 lux, 80-300 lux and less than 80 lux. The comparison and analysis of pictures under different light intensities show that the recognition degree of human eyes to the target is quite different under different light intensities, and the recognition rate of human eyes to the target is higher under 1,000-4,000 lux, 4,000-8,000 lux and 8,000-10,000 lux.

[0063] In step S200B, the morphological color values corresponding to the dynamic light intensity data are extracted, and the morphological color values are input into the trained preference multi-target module as preference data for affinity clustering analysis and training. The morphological color values include RGB color values and morphological values. The preference data includes preference recognition data and preference levels.

[0064] The color and morphological parameters of the pictures of Camellia oleifera fruits under different light intensities are extracted to form preference data, and a SQL-based color and morphological parameter database of Camellia oleifera fruits under different light intensities is established. These parameters will be used as preference data for subsequent algorithms and will be called in the subsequent recognition process.

[0065] Specifically, the morphological characteristics of Camellia oleifera fruits are similar to circles, and the color characteristics are green or light brown. The RGB component value interval is mainly distributed in the interval of 80-210, which is quite different from the color and morphological characteristics of external impurities. Therefore, this study selects length, width, area and perimeter as the basis, and selects circularity, elongation and circularity rate as the indicators to describe the circular morphological characteristics of Camellia oleifera fruits. Meanwhile, the RGB three-dimensional interval image of Camellia oleifera fruits is selected as the color reference object.

[0066] The above 6 parameters are used as indicators to distinguish Camellia oleifera fruits from other targets. These parameters are input into the immune algorithm as preference data, which can better improve the recognition efficiency and speed up the recognition speed. Therefore, this paper selects the color and morphological characteristic parameters of Camellia oleifera fruits under 8 different light intensities as typical preference data, and inputs them into the immune algorithm for target recognition.

[0067] In order to better observe the distribution of color shape characteristics, the color data of camellia oleifera fruits under four different light intensities of 49000 lux, 9500 lux, 4500 lux and 330 lux are selected as preference data for observation (the parameter is the main preference data interval), and each group of data is selected from 10 fruit images.

[0068] Further, the dynamic light self-adaptive adjustment camellia multi-target intelligent detection method provided by the embodiment comprises the following steps:

[0069] In step S110, the to-be-detected camellia image is subjected to region segmentation preprocessing, and the to-be-detected camellia image is segmented into a plurality of regions, and each region corresponds to a matched preference level.

[0070] Due to external interference, branch and leaf overlap and other factors, the light intensity of each region range is different, for easy processing, the light intensity of each region range is taken as The whole image is divided into a plurality of regions, and each region is adapted to a preference level for comprehensive detection.

[0071] It can be seen from the collected image that although the preference algorithm is set at the same time, due to the image collection angle or shielding problem, the light intensity of the camellia tree region is also different when the image range is large, which is easy to cause misidentification; and if the image range is too small, the actual availability of the algorithm is poor. In order to balance the speed and recognition rate requirements, the appropriate collection distance processing is set in this paper, that is, the image collection camera is about 2.3 to 2.5 meters away from the trunk, and on this basis, the pixel The input picture is subjected to region segmentation preprocessing. The picture is segmented into a plurality of regions, and each region corresponds to a matched preference level.

[0072] In step S120, it is judged whether the size of the remaining image after segmentation of the to-be-detected camellia image meets the preset segmentation threshold value, if the size of the remaining image after segmentation of the to-be-detected camellia image is less than the preset segmentation threshold value, the remaining image after segmentation of the to-be-detected camellia image is merged into the adjacent region.

[0073] It is judged whether the size of the remaining image after segmentation of the to-be-detected camellia image meets the preset segmentation threshold value, if the size of the remaining image after segmentation of the to-be-detected camellia image is less than the preset segmentation threshold value, the remaining image after segmentation of the to-be-detected camellia image is merged into the adjacent region.

[0074] Specifically, the camellia image is segmented into 9 regions, and the pixel size of each region is For each region, the corresponding oil tea fruit preferred shape color data in the database is called to match, and each region is calculated separately. When the picture size is large, multi-region joint operation is performed, and multi-thread parallel function is called in software programming to improve the operation speed.

[0075] Further, the dynamic light adaptive adjustment camellia multi-target intelligent detection method provided by the embodiment, in step S200, the affinity of each region of the camellia fruit target is clustered according to the immune affinity formula, the camellia fruit target is recognized from the data, and the coordinate value is returned. The immune affinity formula is:

[0076] (1)

[0077] In formula (1), is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set to the input camellia fruit R component, G component, B component, elongation, circularity and roundness 6 preferred characteristic parameters; is the camellia image to be detected.

[0078] In multi-objective optimization, multiple Pareto optimal solutions are obtained in one run. The expression of the multi-objective formula is:

[0079]

[0080] (2)

[0081] In formula (2), is the target vector, is the maximum / minimum value extracted from the multi-mathematical model, is the function expression of the multi-problem mathematical model to be solved, is the parameter condition expression annotation, is the decision vector, is the decision space, is the target vector, is the target space. As can be seen from the multi-objective formula, the multi-objective optimization problem designs multiple optimization objectives that are contradictory and incomparable to each other. Generally, there is no ideal global optimal solution that can make multiple optimization objectives optimal at the same time. The preferred multi-objective optimization introduces a preference relationship model on the basis of traditional multi-objective optimization. In the preference relationship model, the decision maker needs to give several parameters such as starting point, ending point, indifference threshold vector, preference threshold vector and veto threshold vector.

[0082] In the target recognition of oil tea, a preferred direction can be approximately understood as a dynamic light intensity. The current non-dominated solution closest to the preferred direction is called the most preferred solution, also known as the intermediate antibody, which is used to measure whether other antibodies meet the preferred model. Then all the intermediate antibodies of the most preferred solution can be regarded as the Pareto optimal solution set that meets different light conditions, as shown by the blue line in Figure 2 , and the red line area is the optimal solution set under a certain light intensity. The weighted Chebyshev formula can be defined as follows:

[0083] Suppose is any feasible antibody of the multi-objective optimization shown in formula (2), is the given starting point, is the given end point, satisfying , then the following definitions are given:

[0084] (3)

[0085] (4)

[0086] In formulas (3)~(4), is the multi-problem solution corresponding to the antibody, is the starting point corresponding to the multi-problem solution, is the end point corresponding to the multi-problem solution, is the set terminal value amplification coefficient, is defined as a very small positive real number, and the defined value is For each non-dominated antibody of the current generation, the weighted Chebyshev formula of formula (3) can be used to obtain the measurement value of its distance from the preferred direction , The antibody with the smallest value is the intermediate antibody.

[0087] The intermediate antibody of the non-dominated antibody obtained according to formula (3) is defined as the intermediate antibody of the current non-dominated antibody set , and its preferred level under the dynamic light of the natural environment is defined as:

[0088] (5)

[0089] In formula (5), is the preferred level, is the set terminal value amplification coefficient, , is the multi-problem solution corresponding to the antibody. The defined preferred level can reflect the distance of the antibody from the intermediate antibody.

[0090] The non-dominated antibodies can be effectively distinguished according to the preference levels, so as to facilitate the selection of the antibodies. Meanwhile, the antibodies in the preference region have higher priority, which is beneficial to the algorithm to search in a limited region and reduce the search region of the Pareto front. For a given multi-preference direction problem, the preference level value of an antibody is the minimum level value in all preference directions.

[0091] The SQL database according to the preference levels is shown in Table 1. In the database, each preference level corresponds to a corresponding light intensity, and under the light intensity, the oil tea fruit RGB color component values, and the elongation, circularity and roundness parameter values correspond respectively.

[0092] Table 1: Preference level setting

[0093]

[0094] As can be seen from Table 1, the shape parameters of the oil tea fruit change little under each light intensity, because the shape is not greatly affected by light. However, the color parameter values change greatly due to the influence of external light. Moreover, with the increase of brightness, the overall RGB component values increase, which is consistent with the actual situation under dynamic light in natural environment.

[0095] Further, the oil tea multi-target intelligent detection method provided by the embodiment is dynamically adjusted under dynamic light. In step S300, based on the light intensity analysis result, the oil tea antibody population is selected, crossed and mutated to obtain a new oil tea population. The cell affinity in the new oil tea population is calculated as follows:

[0096] (6)

[0097] (7)

[0098] In formulas (6) and (7), is the cell new cell formed after mutation, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is an exponential decay variable in the adjustment function, is the cell self-adaptive value after standardization processing, is the self-adjusting mutation coefficient. The cell affinity in the new population is calculated. If the cell affinity is less than the affinity threshold, it is retained. Otherwise, the cell is inhibited.

[0099] Another aspect of the present application relates to a dynamic light adaptive adjustment camellia multi-target intelligent detection system, comprising a generation module, an analysis module and an acquisition module, wherein the generation module is configured to perform regional segmentation preprocessing on the to-be-detected camellia image to generate a multi-region segmented camellia image; the analysis module is configured to input the multi-region segmented camellia image into a camellia fruit multi-target recognition immune algorithm, and perform light intensity analysis on the multi-region segmented camellia image by a trained preference multi-target module; the preference multi-target module is configured to perform target recognition on the morphological color values of the camellia fruits under different light conditions as preference data; and the acquisition module is configured to acquire the recognition result of the camellia fruits in the to-be-detected camellia image according to the light intensity analysis result output by the trained preference multi-target module.

[0100] Further, the dynamic light adaptive adjustment camellia multi-target intelligent detection system provided by the embodiment further comprises a collection module and an extraction module, wherein the collection module is configured to collect dynamic light intensity data of the camellia fruits under different light conditions, including different weather and different time period light intensities; and the extraction module is configured to extract morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into a to-be-trained preference multi-target module for affinity clustering analysis training, wherein the morphological color values include RGB color values and morphological values; and the preference data includes preference recognition data and preference levels.

[0101] Further, the dynamic light adaptive adjustment camellia multi-target intelligent detection system provided by the embodiment further comprises a collection module and an extraction module, wherein the collection module is configured to collect dynamic light intensity data of the camellia fruits under different light conditions, including different weather and different time period light intensities; and the extraction module is configured to extract morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into a to-be-trained preference multi-target module for affinity clustering analysis training, wherein the morphological color values include RGB color values and morphological values; and the preference data includes preference recognition data and preference levels.

[0102] Further, in the analysis module, the affinity clustering results of the camellia fruit targets in each region are integrated according to an immune affinity formula to identify the camellia fruit targets from the data and return coordinate values, and the immune affinity formula is:

[0103] (8)

[0104] In formula (8), is an affinity value, is a Gaussian function symbol, is a Gaussian function parameter, The six preference characteristic parameters of the input Camellia oleifera R component, G component, B component, elongation, circularity and roundness are set; The Camellia oleifera image to be detected is obtained.

[0105] Further, the embodiment provides a Camellia oleifera multi-target intelligent detection system with dynamic light adaptation, in the acquisition module, based on the light intensity analysis result, the Camellia oleifera antibody population is selected, crossed and mutated to obtain a new Camellia oleifera population, and the cell affinity is calculated in the new Camellia oleifera population, and the cell affinity is:

[0106] (9)

[0107] (10)

[0108] In formulas (9) and (10), is a cell a new cell formed after mutation, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is an exponential decay variable in the adjustment function, is the cell adaptive value after standardization processing, is the self-adjusting mutation coefficient.

[0109] Compared with the prior art, the Camellia oleifera multi-target intelligent detection method and system with dynamic light adaptation provided by the embodiment perform regional segmentation preprocessing on the Camellia oleifera image to be detected to generate a multi-region segmented Camellia oleifera image; input the multi-region segmented Camellia oleifera image into a Camellia oleifera fruit multi-target recognition immune algorithm, and perform light intensity analysis on the multi-region segmented Camellia oleifera image by the trained preference multi-target module; wherein the preference multi-target module is used for target recognition on the shape and color values of the Camellia oleifera fruit under different light intensities as preference data; and according to the light intensity analysis result output by the trained preference multi-target module, the recognition result of the Camellia oleifera fruit in the Camellia oleifera image to be detected is obtained. The Camellia oleifera multi-target intelligent detection method and system with dynamic light adaptation provided by the embodiment have the following beneficial effects:

[0110] 1) The recognition rate is improved by optimizing the multi-target preference data, the preference data corresponds to six features of the shape and color of the Camellia oleifera fruit under different light intensities, and the problem of dynamic change of the target recognition rate under uneven light intensity is solved.

[0111] 2) The efficiency of the immune algorithm is improved by improving and optimizing the preference level, image partition and other immune algorithms, and the efficient algorithm structure of the immune algorithm is used to make it applicable to the Camellia oleifera natural picking environment with high real-time rate requirement and low-cost software and hardware system.

[0112] 3) By comparing with 5 kinds of algorithms such as classic-AIS, classic genetic algorithm (GA), Faster R-CNN, BP neural network (BP-Network) and YOLOv8n, the recognition rate of the algorithm proposed in this paper under natural light reaches 93.8%, and the recognition time is 160 ms level, which proves the availability of the algorithm under dynamic light.

[0113] The test and test of the model in this embodiment are based on more than 4500 test images under different light environment conditions, and mainly researches the identification of the color and morphological characteristics of Camellia oleifera fruits. The dynamic light recognition not only guarantees the efficiency but also improves the speed, and provides an effective solution for Camellia oleifera fruit recognition in natural environment.

[0114] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to the embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A dynamic light adaptive adjustment of camellia oleifera multi-objective intelligent detection method, characterized by, The method comprises the following steps: The image of the oil tea to be detected is subjected to region segmentation preprocessing to generate a multi-region segmented oil tea image; Dynamic light intensity data of the oil tea fruit under different illuminations are collected, wherein the different illuminations include light intensities of different weathers and different time periods; Morphological color values corresponding to the dynamic light intensity data are extracted, and the morphological color values are input as preference data into a trained preference multi-objective module for affinity clustering analysis training, wherein the morphological color values include RGB color values and morphological values; the preference data include preference recognition data and preference levels; The multi-region segmented oil tea image is input into an oil tea fruit multi-objective recognition immune algorithm, and typical preference recognition data and preference levels of morphological colors corresponding to the dynamic light intensity data detected by the luxmeter are fused with a region segmentation module by the trained preference multi-objective module to analyze the light intensity of the multi-region segmented oil tea image; wherein the preference multi-objective module is used for target recognition of morphological color values of the oil tea fruit under different illuminations as preference data; An identification result of the oil tea fruit in the image to be detected is obtained according to a light intensity analysis result output by the trained preference multi-objective module.

2. The dynamic light adaptive adjustment of oil tea multi-objective intelligent detection method of claim 1, wherein, The step of subjecting the image of the oil tea to be detected to region segmentation preprocessing to generate a multi-region segmented oil tea image comprises: The image of the oil tea to be detected is subjected to region segmentation preprocessing, and the image of the oil tea to be detected is segmented into a plurality of regions, each region corresponding to a matched preference level; It is judged whether the size of the remaining image after the image of the oil tea to be detected is segmented meets a preset segmentation threshold value, if the size of the remaining image after the image of the oil tea to be detected is segmented is less than the preset segmentation threshold value, the remaining image after the image of the oil tea to be detected is segmented is merged into a neighboring region. 3.The dynamic light self-adaptive adjustment of camellia oleifera multi-objective intelligent detection method of claim 1, wherein, In the step of inputting the multi-region segmented oil tea image into an oil tea fruit multi-objective recognition immune algorithm and analyzing the light intensity of the multi-region segmented oil tea image by the trained preference multi-objective module, affinity clustering results of oil tea fruit targets of each region are comprehensively calculated according to an immune affinity formula to identify the oil tea fruit targets from data and return coordinate values, wherein the immune affinity formula is: ； wherein, is an affinity value, is a Gaussian function symbol, is a Gaussian function parameter, Set the input Camellia fruit R component, G component, B component, elongation, circularity and perfection 6 preferred characteristic parameters; is a to-be-detected Camellia image. 4.The dynamic light self-adaptive adjustment of camellia oleifera multi-objective intelligent detection method of claim 1, wherein, In the step of obtaining an identification result of the oil tea fruit in the image to be detected according to a light intensity analysis result output by the trained preference multi-objective module, a new oil tea population is obtained by selecting, crossing and mutating an oil tea population based on the light intensity analysis result, and cell affinity is calculated in the new oil tea population, wherein the cell affinity is: ； wherein, is a cell new cell formed after mutation, is a primary cell, is a Gaussian random variable with mean 0 and deviation 1; is an exponential decay variable in the regulation function, is a cell self-adaptive value after standardization, is a self-regulated mutation coefficient.

5. A dynamic light adaptive adjustment of Camellia oleifera multi-objective intelligent detection system, characterized by, It comprises: A generation module is configured to subject the image of the oil tea to be detected to region segmentation preprocessing to generate a multi-region segmented oil tea image; A collection module is configured to collect dynamic light intensity data of the oil tea fruit under different illuminations, wherein the different illuminations include light intensities of different weathers and different time periods; An extraction module is configured to extract morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into a trained preference multi-objective module for affinity clustering analysis training, wherein the morphological color values include RGB color values and morphological values; the preference data include preference recognition data and preference levels; The analysis module is configured to input the multi-region segmented oil tea image into an oil tea fruit multi-target recognition immune algorithm, and fuse morphological color typical preference recognition data corresponding to dynamic light intensity data detected by the luxmeter, a preference level and a region segmentation module by a trained preference multi-target module to analyze the light intensity of the multi-region segmented oil tea image; the preference multi-target module is configured to recognize the morphological color values of the oil tea fruits under different illuminations as preference data. The acquisition module is configured to acquire a recognition result of the oil tea fruits in the to-be-detected oil tea image according to the light intensity analysis result output by the trained preference multi-target module. 6.The dynamic light self-adaptive adjustment of camellia oleifera multi-objective intelligent detection system of claim 5, wherein, The generation module includes: The segmentation unit is configured to perform region segmentation preprocessing on the to-be-detected oil tea image, and divide the to-be-detected oil tea image into a plurality of regions, each region corresponding to a matched preference level. The judgment unit is configured to judge whether the size of the remaining image after the to-be-detected oil tea image is segmented meets a preset segmentation threshold value, and if the size of the remaining image after the to-be-detected oil tea image is segmented is less than the preset segmentation threshold value, the remaining image after the to-be-detected oil tea image is segmented is merged into a neighboring region. 7.The dynamic light self-adaptive adjustment of camellia oleifera multi-objective intelligent detection system of claim 5, wherein, In the analysis module, affinity clustering results of the oil tea fruit targets in each region are integrated according to an immune affinity formula to identify the oil tea fruit targets from the data and return coordinate values, and the immune affinity formula is: ； wherein, is an affinity value, is a Gaussian function symbol, is a Gaussian function parameter, Set the input Camellia fruit R component, G component, B component, elongation, circularity and roundness 6 item preference characteristic parameters; is a to-be-detected Camellia image. 8.The dynamic light self-adaptive adjustment of camellia oleifera multi-objective intelligent detection system of claim 5, wherein, In the acquisition module, an oil tea population is selected, crossed and mutated based on the light intensity analysis result to obtain a new oil tea population, and cell affinity is calculated in the new oil tea population, and the cell affinity is: ； wherein, is a cell new cell formed after variation, is a primary cell, is a Gaussian random variable with mean 0 and deviation 1; is an exponential decay variable in the regulation function, is a cell self-adaptive value after standardization, is a self-regulated variation coefficient.

Images