Camellia oleifera multi-target intelligent detection method and system based on dynamic illumination adaptive adjustment

Abstract

The invention discloses a camellia oleifera multi-target intelligent detection method and system based on dynamic illumination adaptive adjustment, and the method comprises the steps: carrying out the region segmentation preprocessing of a to-be-detected camellia oleifera image, and generating a multi-region segmented camellia oleifera image; inputting the multi-region segmented camellia oleifera image into a camellia oleifera fruit multi-target recognition immune algorithm, and performing light intensity analysis on the multi-region segmented camellia oleifera image by a trained preference multi-target module; wherein the preference multi-target module is used for carrying out target identification on form color values of camellia oleifera fruits under different illumination conditions as preference data; and according to a light intensity analysis result output by the trained preference multi-target module, obtaining a recognition result of the camellia oleifera fruit in the camellia oleifera image to be detected. According to the method, the efficiency is ensured and the speed is increased in dynamic illumination recognition, and an effective solution is provided for camellia oleifera fruit recognition in a natural environment.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular discloses a multi-target intelligent detection method and system for dynamic light adaptability adjustment of oil-tea camellia. Background Technique

[0002] Currently, in the process of the automation field of oil-tea camellia production and processing, target visual recognition, especially visual recognition under natural light, is a very important link. Only by maximizing the recognition of target oil-tea camellia fruits in the natural environment is it beneficial for the picking machine to perform subsequent unmanned automated picking, thereby saving production and processing costs.

[0003] In recent years, many scholars at home and abroad have conducted a large number of studies on intelligent algorithms, which have also been widely applied in the field of agricultural and forestry target recognition. Specifically in the field of intelligent target recognition of oil-tea camellia, a large number of studies have also been carried out in recent years. Chen Fengjun et al. proposed an improved YOLOv7 algorithm to detect the maturity of oil-tea camellia fruits, introduced a cross-shaped cross-attention mechanism to optimize the backbone network, and improved the intersection over union. After training, it was measured that the average detection time was 0.77 s, proving the usability of the algorithm. Xie Weijun et al. used AlexNet as the backbone network to solve the problem of identifying broken seeds in existing oil-tea camellia color sorters, and optimized the number of layers of its subsequent network, and finally completed the color sorting recognition of the integrity of oil-tea camellia broken seeds. Compared with traditional methods, the recognition rate has been greatly improved. Zhou Hongping et al. used the YOLOv8n recognition algorithm to classify and recognize oil-tea camellia fruits. This method uses the coco dataset as the training model, combines four factors such as the learning method, the amount of training data, the learning rate, and the number of training rounds, and conducted 52 ablation experiments on YOLOv8n, and finally achieved better results; Meng Zhichao et al. proposed an improved VGG16 network for the problem of chaotic varieties of oil-tea camellia seedlings in the market and completed training and testing on the oil-tea camellia leaf dataset; Wang Jinpeng et al. used the SCConv as the backbone network YOLOv8n algorithm to identify the problem of low recognition rate caused by the occlusion phenomenon of oil-tea camellia; Duan Yufei et al. conducted research on the visual sorting of fruit shells and seeds of oil-tea camellia shelling machines, optimized with VGG16 as the backbone network, alleviated the problem of gradient explosion in image detection, and thus improved the system robustness.

[0004] Many scholars in the field of domestic and foreign agriculture and forestry have conducted a lot of research on light and occlusion recognition that is greatly affected by light. Liu et al. used an improved Faster R-CNN method for the recognition of multiple clusters of green persimmons under different light conditions in natural environments. YANG et al. enhanced low-light images by combining two CNNs to achieve image reconstruction. Du Jinzhi et al. used a low-light enhancement method to solve the light problem. The method was to improve the GAN to enhance the kiwifruit flower images under low-light conditions, thus achieving the brightness enhancement of kiwifruit flower images under low-light conditions and effectively suppressing the appearance of noise. Li Xu et al. added a fusion of efficient channel CA (coordinate attention) attention mechanism to the YOLOX algorithm module to improve the ability of the algorithm to recognize 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 tomato surface and then used the Otsu method to extract the surface coloring area, and further used the local brightness equalization method to repair and solve the high-brightness reflection on the tomato surface caused by light changes. Finally, the average correct rate of the coloring area on the tomato surface reached 92.96%. Wang Chuanyu et al. proposed a dynamic light removal method based on multi-exposure image fusion. After calculating the light radiation intensity of the corn canopy, the method mapped different light radiation intensities to the RGB space, and the brightness of the hemisphere image was evenly distributed after mapping, thus eliminating the influence of light changes on the image. Bao Wenxia et al. used the Retinex algorithm to enhance the wheat images of unmanned aerial vehicles. And a coordinate attention mechanism was added to the backbone network of YOLOv5 recognition to refine the model features. After experiments, the accuracy was increased by an average of 4.1% compared with the basic recognition algorithm.

[0005] Although the above methods have achieved certain results in the field of detecting agricultural and forestry targets and light, they do not perform targeted light optimization for the dynamically changing external environment in nature. The picking time of oil tea is generally between September and November every year. At present, oil tea has been basically automated in the picking and processing fields, but there are still many problems in the automated recognition process. Mainly in the natural environment, the light conditions in the oil tea picking base are complex. Affected by weather and the occlusion of branches and leaves, the light has a great impact on the recognition rate. At the same time, the recognition rates in different oil tea tree areas are also different. This leads to different recognition rates in the morning, noon and evening after debugging with the same recognition algorithm, and the recognition rate is high in a certain area of the oil tea tree while low in another area. This dynamic light intensity change in the natural environment brings great difficulties to intelligent picking recognition. Summary of the Invention

[0006] The present invention provides a method and system for intelligent detection of multiple oil tea targets with adaptive adjustment of dynamic light, aiming to solve at least one defect existing in the above-mentioned prior art.

[0007] One aspect of the present invention relates to a multi-object intelligent detection method for oil-tea camellia with dynamic light-adaptive adjustment, comprising the following steps: Preprocess the oil-tea camellia image to be detected by region segmentation to generate a multi-region segmented oil-tea camellia image; Input the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits, and the trained preference multi-object module performs light intensity analysis on the multi-region segmented oil-tea camellia image; wherein, the preference multi-object module is used to perform target recognition on the morphological color values of oil-tea camellia fruits under different illuminations as preference data; Obtain the recognition result of the oil-tea camellia fruits in the oil-tea camellia image to be detected according to the light intensity analysis result output by the trained preference multi-object module.

[0008] Further, before the step of inputting the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits and the trained preference multi-object module performing light intensity analysis on the multi-region segmented oil-tea camellia image, it includes: Collect the dynamic light intensity data of oil-tea camellia fruits under different illuminations, and different illuminations include the light intensities in different weathers and different time periods; Extract the morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into the preference multi-object module to be trained 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.

[0009] Further, the step of preprocessing the oil-tea camellia image to be detected by region segmentation to generate a multi-region segmented oil-tea camellia image includes: Preprocess the oil-tea camellia image to be detected by region segmentation, divide the oil-tea camellia image to be detected into several regions, and each region corresponds to a matching preference level; Judge whether the size of the remaining image after segmentation of the oil-tea camellia image to be detected meets the preset segmentation threshold. If the size of the remaining image after segmentation of the oil-tea camellia image to be detected is smaller than the preset segmentation threshold, incorporate the remaining image after segmentation of the oil-tea camellia image to be detected into the adjacent region.

[0010] Further, in the step of inputting the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits and the trained preference multi-object module performing light intensity analysis on the multi-region segmented oil-tea camellia image, according to the immune affinity formula, comprehensively combine the affinity clustering results of the oil-tea camellia fruit targets in each region, identify the oil-tea camellia fruit targets from the data and return the coordinate values. The immune affinity formula is:

[0011] Wherein, is the affinity value, is the Gaussian function symbol, is the parameter of the Gaussian function, which is set as the six preference characteristic parameters of the input R component, G component, B component, elongation, roundness, and perfection degree of the oil-tea fruit; is the oil-tea image to be detected.

[0012] Furthermore, in the step of obtaining the recognition result of the oil-tea fruit in the oil-tea image to be detected based on the light intensity analysis result output by the trained preference multi-object module, based on the light intensity analysis result, selection, crossover, and mutation operations are performed on the oil-tea antibody population to obtain a new oil-tea population, and the cell affinity is calculated in the new oil-tea population. The cell affinity is:

[0013] where, is the new cell formed after the mutation of cell , is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is the exponential decay variable in the adjustment function, is the cell adaptive value after normalization processing, is the self-adjusting mutation coefficient.

[0014] On the other hand, the present invention relates to an oil-tea multi-object intelligent detection system with dynamic light intensity adaptive adjustment, including: a generation module, configured to perform region segmentation preprocessing on the oil-tea image to be detected and generate a multi-region segmented oil-tea image; an analysis module, configured to input the multi-region segmented oil-tea image into the oil-tea fruit multi-object recognition immune algorithm, and perform light intensity analysis on the multi-region segmented oil-tea image by the trained preference multi-object module; the preference multi-object module is used to perform target recognition on the morphological color values of the oil-tea fruit under different light intensities as preference data; an acquisition module, configured to obtain the recognition result of the oil-tea fruit in the oil-tea image to be detected according to the light intensity analysis result output by the trained preference multi-object module.

[0015] Furthermore, the oil-tea multi-object intelligent detection system with dynamic light intensity adaptive adjustment further includes: a collection module, configured to collect the dynamic light intensity data of the oil-tea fruit under different light intensities, where different light intensities include the light intensities in different weathers and different time periods; an extraction module, configured to extract the morphological color values corresponding to the dynamic light intensity data and input the morphological color values as preference data into the preference multi-object module to be trained 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.

[0016] Further, the generation module includes: A segmentation unit, configured to perform regional segmentation preprocessing on the oil-tea camellia image to be detected, and segment the oil-tea camellia image to be detected into several regions, and each region corresponds to a matching preference level; A judgment unit, configured to judge whether the size of the remaining image after segmentation of the oil-tea camellia image to be detected meets a preset segmentation threshold. If the size of the remaining image after segmentation of the oil-tea camellia image to be detected is smaller than the preset segmentation threshold, the remaining image after segmentation of the oil-tea camellia image to be detected is incorporated into an adjacent region.

[0017] Further, in the analysis module, according to the immune affinity formula, the affinity clustering results of the oil-tea camellia fruit targets in each region are synthesized, and the oil-tea camellia fruit targets are identified from the data and the coordinate values are returned. The immune affinity formula is:

[0018] Wherein, is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set as 6 preference feature parameters of the input R component, G component, B component, elongation, circularity and roundness of the oil-tea camellia fruit; is the oil-tea camellia image to be detected.

[0019] Further, in the acquisition module, based on the light intensity analysis result, selection, crossover and mutation operations are performed on the oil-tea camellia antibody population to obtain a new oil-tea camellia population, and the cell affinity is calculated in the new oil-tea camellia population. The cell affinity is:

[0020] Wherein, is the cell formed after mutation of the new cell, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is the exponential decay variable in the adjustment function, is the cell adaptive value after being standardized, is the self-adjusting mutation coefficient.

[0021] The beneficial effects achieved by the present invention are: The present invention provides a method and system for intelligent multi-target detection of oil-tea camellia with dynamic light-adaptive adjustment. The oil-tea camellia image to be detected is preprocessed by region segmentation to generate a multi-region segmented oil-tea camellia image. The multi-region segmented oil-tea camellia image is input into an immune algorithm for multi-target recognition of oil-tea camellia fruits, and the trained preference multi-target module performs light intensity analysis on the multi-region segmented oil-tea camellia image. Among them, the preference multi-target module is used to perform target recognition on the morphological and color values of oil-tea camellia fruits under different illuminations as preference data. According to the light intensity analysis result output by the trained preference multi-target module, the recognition result of the oil-tea camellia fruits in the oil-tea camellia image to be detected is obtained. The method and system for intelligent multi-target detection of oil-tea camellia with dynamic light-adaptive adjustment provided by the present invention have the following beneficial effects: 1) The recognition rate is improved by optimizing and improving with multi-target preference data. The preference data corresponds to 6 features of the morphology and color of oil-tea camellia fruits under different light intensities, and solves the problem of dynamic change of the target recognition rate under uneven light intensity.

[0022] 2) By improving and optimizing immune algorithms such as preference level and image partitioning, the efficiency of the immune algorithm is improved. Using the efficient algorithm structure of the immune algorithm, it is applicable to the natural picking environment of oil-tea camellia with high real-time requirements and low-cost software and hardware systems.

[0023] 3) By comparing with 5 algorithms, namely the traditional immune algorithm (Classic-AIS), the traditional genetic algorithm (GA), Faster R-CNN, BP neural network (BP-Network), and YOLOv8n, the recognition rate of the algorithm proposed in this paper reaches up to 93.8% under natural light, and the recognition time is at the 160ms level, which proves the availability of the algorithm proposed in this paper under dynamic light.

[0024] The experiments and tests of the research model of the present invention are based on more than 4,500 test images under different light environment conditions, and mainly focus on the research of the preference feature recognition of the color and morphological characteristics of oil-tea camellia fruits. It ensures both efficiency and speed in dynamic light recognition, and provides an effective solution for the recognition of oil-tea camellia fruits in natural environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic flowchart of an embodiment of a method for intelligent multi-target detection of oil-tea camellia with dynamic light-adaptive adjustment according to the present invention; Figure 2 It is a schematic diagram of the solution set of oil-tea camellia target recognition in a method for intelligent multi-target detection of oil-tea camellia with dynamic light-adaptive adjustment according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] To better understand the above technical solution, the following will provide a detailed description of the above technical solution in combination with the accompanying drawings of the specification and specific implementation manners.

[0027] As Figure 1 shown, the first embodiment of the present invention proposes a multi-object intelligent detection method for camellia oleifera with dynamic light intensity self-adaptive adjustment, including the following steps: Step S100: Perform regional segmentation preprocessing on the camellia oleifera image to be detected to generate a multi-region segmented camellia oleifera image.

[0028] Perform regional segmentation on the camellia oleifera image to be detected, select pixels for regional segmentation detection, and the number of regions can be adjusted according to the camera focal length and the distance of acquisition.

[0029] Step S200: Input the multi-region segmented camellia oleifera image into the multi-object recognition immune algorithm for camellia oleifera fruits, and perform light intensity analysis on the multi-region segmented camellia oleifera image by the trained preference multi-object module; among them, the preference multi-object module is used to perform target recognition on the morphological color values of camellia oleifera fruits under different light intensities as preference data.

[0030] Input the generated multi-region segmented camellia oleifera image into the multi-object recognition immune algorithm for camellia oleifera fruits for affinity clustering analysis. The preference multi-object module in the multi-object recognition immune algorithm for camellia oleifera fruits is used to fuse the typical preference recognition data, preference levels of morphological colors corresponding to the dynamic light intensity data detected by the illuminometer with the region segmentation module to identify camellia oleifera fruits under different light intensities.

[0031] Step S300: Obtain the recognition result of camellia oleifera fruits in the camellia oleifera image to be detected according to the light intensity analysis result output by the trained preference multi-object module.

[0032] After inputting preference data into each segmentation region, comprehensively combine the affinity clustering results of camellia oleifera fruit targets in each region, identify the camellia oleifera fruit targets and return the coordinate values.

[0033] Furthermore, for the multi-object intelligent detection method for camellia oleifera with dynamic light intensity self-adaptive adjustment provided in this embodiment, before step S200, it includes: Step S200A: Collect the dynamic light intensity data of camellia oleifera fruits under different light intensities, where different light intensities include the light intensities in different weather and different time periods.

[0034] The acquisition device is a SONY ZV-E10L mirrorless digital camera, and the camera is equipped with a professional lens Sigma 18-50 to obtain better image quality. The camellia oleifera variety collected is Xianglin-1 camellia oleifera.

[0035] The sampling time was on sunny and cloudy days. Camellia oleifera pictures were collected from 9 am to 9 pm under natural light. Collection was carried out approximately every hour according to the light conditions, for a total of 12 collection time points. On average, about 570 pictures were collected for each light intensity for subsequent comprehensive analysis. The illuminance range of the pictures was from 50 lux to 100,000 lux, and the Camellia oleifera fruits in the natural environment were the targets to be collected.

[0036] Analysis of the target images of Camellia oleifera fruits after collection shows that the brightness of the pictures of Camellia oleifera fruits growing in the natural environment is not only greatly affected by the light angle, but also the mutual occlusion between tree branches, leaves and fruits will affect the picture quality. This complex natural environment brings great difficulties to the practical application of the forestry picking vision system, and it is also a problem that must be solved for the automation of the Camellia oleifera vision system.

[0037] The illuminometer is also called a luxmeter, which is an instrument used to measure the light intensity, measuring the degree to which the surface of an object is illuminated, that is, the ratio of the luminous flux to the illuminated area. The light intensity measurement device selected for this experiment is the Delixi LSK-2304 high-precision illuminometer, with an illuminance range of 0-200,000 lux (lux), a resolution of 1 lux, an operating temperature environment of 0-40 degrees Celsius, equipped with a high-precision photosensitive probe, with a temperature detection function, and the illuminance measurement accuracy is plus or minus 4%, supporting the storage of historical illuminance data. Its data indicators are applicable to the dynamic natural light environment of Camellia oleifera. When measuring, the tripod needs to be fixed towards the image acquisition direction to obtain stable light intensity data.

[0038] Considering the influence of weather on light, the measurement time was divided into sunny and cloudy environments, and 9:00-21:00 of the two weather conditions were selected to measure the light intensity. Since the light intensity is dynamically changing, the illuminance was measured for two minutes each time, and the average value was comprehensively taken with a discrete value every 10 seconds.

[0039] The most influential factor on the target detection of Camellia oleifera fruits growing in the natural environment is the light intensity. The recognition rate of pictures is not high under the relatively dark light intensity at noon on a normal sunny day or in the early morning and evening on a cloudy day, while the data indicators at noon on a cloudy day are similar to those in the early morning and evening on a sunny day, and the recognition rate is relatively high. More than 4,500 pictures collected under different light intensities were analyzed. Finally, the pictures were divided into 8 levels according to the illuminance value: above 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 below 80 lux for separate processing. At the same time, through the comparative analysis of pictures in different light intensity ranges, it can be seen that there are significant differences in the human eye's recognition of targets under different light intensities. Among them, the human eye recognition rate of targets is relatively high under the three light intensities of 1,000-4,000 lux, 4,000-8,000 lux, and 8,000-10,000 lux.

[0040] Step S200B: Extract the morphological color values corresponding to the dynamic light intensity data, and use the morphological color values as preference data to input into the to-be-trained multi-objective preference module 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.

[0041] Extract the color and morphological parameters of the collected pictures of camellia oleifera fruits under different lights to form preference data, and establish a database of color and morphological parameters of camellia oleifera fruits under different lights based on SQL. These parameters will be used as the preference data for subsequent algorithms and called during subsequent recognition processes.

[0042] Specifically, the morphological characteristics of camellia oleifera fruits belong to the shape of a roundish circle, and the color characteristics satisfy cyan or light brown. The RGB component value range is mainly distributed in the range of 80 - 210, which is quite different from that of external impurities in terms of color and morphological characteristics. Therefore, in this study, based on length, width, area, and perimeter, roundness, elongation, and roundness rate are selected as three indicators to describe the roundish morphological characteristics of camellia oleifera fruits. At the same time, the RGB three-dimensional interval image within the range of camellia oleifera fruits is selected as the color reference object.

[0043] The above six parameters are used as indicators to distinguish the camellia oleifera fruit target from other targets. Inputting these parameters as preference data into the immune algorithm can better improve the recognition efficiency of the algorithm and speed up the recognition speed. Therefore, in this paper, the color and morphological characteristic parameters of camellia oleifera fruits under eight different lights are selected as typical preference data and input into the immune algorithm for target recognition.

[0044] To better observe the distribution of color and morphological characteristics, in this paper, the color data of camellia oleifera fruits under four different light intensities of 49000lux, 9500lux, 4500lux, and 330lux are selected as preference data for observation (the parameters are the main preference data intervals), and each group of data is selected from 10 fruit images.

[0045] Furthermore, for the intelligent multi-objective detection method of camellia oleifera with dynamic light adaptive adjustment provided in this embodiment, step S100 includes: Step S110: Perform region segmentation preprocessing on the camellia oleifera image to be detected, divide the camellia oleifera image to be detected into several regions, and each region corresponds to a matching preference level.

[0046] Due to factors such as external interference and overlap of branches and leaves, the light intensity in each region range is different. For the convenience of processing, Pixels are used as a segmentation measurement region, and the overall image is divided into several regions, and each region is adapted with a preference level for comprehensive detection.

[0047] As can be seen from the acquired images, although the preference algorithm is set within the same time, due to the image acquisition angle or occlusion problems, when the image range is large, the light intensity in the camellia oleifera tree area is also different, which easily leads to the problem of misidentification; while if the image range is too small, the actual usability of the algorithm is poor. To balance the requirements of speed and recognition rate, this paper sets an appropriate acquisition distance processing, that is, the image acquisition camera is about 2.3 to 2.5 meters away from the tree trunk. On this basis, taking pixels as the basic segmentation area, the input picture is first preprocessed by region segmentation. The picture is segmented into several regions, and each region corresponds to a matching preference level.

[0048] Step S120: Determine whether the size of the remaining image after segmentation of the camellia oleifera image to be detected meets the preset segmentation threshold. If the size of the remaining image after segmentation of the camellia oleifera image to be detected is smaller than the preset segmentation threshold, then incorporate the remaining image after segmentation of the camellia oleifera image to be detected into the adjacent region.

[0049] Determine whether the size of the remaining image after segmentation of the camellia oleifera image to be detected meets the preset segmentation threshold. If the remaining size after segmentation does not meet the segmentation size, then set a threshold in the algorithm to incorporate it into the adjacent region. Specifically, if the size of the remaining image after segmentation of the camellia oleifera image to be detected is smaller than the preset segmentation threshold, then incorporate the remaining image after segmentation of the camellia oleifera image to be detected into the adjacent region.

[0050] Specifically, the camellia oleifera image is segmented into 9 regions, and the pixel size of each region is , for each region, call the corresponding preferred morphological color data of camellia oleifera fruits in the database for matching, and perform separate operations on each region. When the picture scale is large, multi-region joint operations are performed, which is implemented by calling the multi-thread parallel function in software programming to improve the operation speed.

[0051] Furthermore, for the multi-object intelligent detection method of camellia oleifera with dynamic light adaptive adjustment provided in this embodiment, in step S200, according to the immune affinity formula, comprehensively cluster the affinity results of the camellia oleifera fruit targets in each region, identify the camellia oleifera fruit targets from the data and return the coordinate values. The immune affinity formula is: (1) In formula (1), is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set as 6 preference feature parameters of the input camellia oleifera fruit R component, G component, B component, elongation rate, circularity and roundness; is the camellia oleifera image to be detected.

[0052] In multi-objective optimization, multiple Pareto optimal solutions are obtained in one run. The expression of the multi-objective formula is:

[0053] (2) In formula (2), is the objective vector, is the maximum / minimum value extracted from multiple mathematical models, is the function expression of the multi-problem mathematical model to be solved for the objective, is the annotation of the parameter condition expression, is the decision vector, is the decision space, is the objective vector, is the objective space. It can be seen from the multi-objective formula that the multi-objective optimization problem involves multiple conflicting and incomparable optimization objectives. Generally speaking, there is no ideal global optimal solution that can simultaneously optimize multiple objectives to be optimized. Preference-based multi-objective optimization introduces a preference relation model on the basis of traditional multi-objective optimization. In the preference relation model, the decision maker needs to give several parameters such as the starting point, the ending point, the indifference threshold vector, the preference threshold vector, and the veto threshold vector respectively.

[0054] In the target recognition of Camellia oleifera, a preference direction can be approximately understood as a kind of dynamic light intensity. The current non-dominated solution closest to the preference direction is called the most preferred solution, also known as the intermediate antibody, which is used to measure whether other antibodies meet the preference model. Then, the intermediate antibodies of all the most preferred solutions can be regarded as the Pareto optimal solution set that satisfies different light conditions, as shown by the blue line in Figure 2 , while the red line area is the optimal solution set under a certain light intensity. It can be defined by the weighted Chebyshev formula as follows: Suppose is any feasible antibody for the multi-objective optimization shown in formula (2), is the given starting point, is the given ending point, satisfying condition, then the following definitions are available: (3) (4) In formulas (3) to (4), is the multi-problem solution corresponding to the antibody, is the starting point corresponding to the multi-problem solution, is the ending point corresponding to the multi-problem solution, is the set final value amplification factor, is defined as a very small positive real number, and the defined value is , for each non-dominated antibody in the current generation, through the weighted Chebyshev formula of formula (3), the metric value of its distance to the preference direction can be obtained , The antibody with the smallest value is the intermediate antibody.

[0055] The intermediate antibody of the non-dominated antibody obtained according to formula (3), for any antibody in the current non-dominated antibody set, for the natural environment under dynamic light, its preference level is defined as: (5) In formula (5), is the preference level, is the set final value amplification factor, , are the multi-problem solutions corresponding to the antibody. The defined preference level can reflect the distance of the antibody from the intermediate antibody.

[0056] According to the preference level, the non-dominated antibodies can be effectively distinguished, which is conducive to the selection of antibodies. At the same time, the antibodies within the preference region have a higher priority, which is conducive to the algorithm to concentrate on searching a limited area and reduce the Pareto front search area. For a given multi-preference direction problem, the preference level value of an antibody is the minimum level value under all preference directions.

[0057] An SQL database is established according to the preference level as shown in Table 1. In the database, each preference level corresponds to the corresponding light intensity, and under this light intensity, it corresponds to the RGB color component values of the oil tea fruit, as well as the parameter values of elongation, roundness, and fullness.

[0058] Table 1 Preference level setting

[0059] As can be seen from Table 1, the morphological parameters of the oil tea fruit under each light intensity change little because its shape is little affected by light, while the color parameter values change greatly due to the influence of external light. Moreover, as the brightness increases, the overall RGB component values increase, which is consistent with the actual situation under dynamic light in the natural environment.

[0060] Furthermore, in the dynamic light adaptive adjustment multi-objective intelligent detection method for oil tea provided in this embodiment, in step S300, based on the light intensity analysis result, selection, crossover, and mutation operations are performed on the oil tea antibody population to obtain a new oil tea population, and the cell affinity is calculated in the new oil tea population. The cell affinity is: (6) (7) In formulas (6) to (7), is the cell formed after cell mutation. is the original cell, which is a Gaussian random variable with a mean of 0 and a deviation of 1; is the exponentially decaying variable in the adjustment function, which is the cell adaptive value after normalization processing, is the self-adjusting coefficient of variation. Calculate the cell affinity in the immune algorithm in the new population. Those less than the affinity threshold are retained, and other cells are inhibited.

[0061] Another aspect of the present invention relates to a multi-objective intelligent detection system for Camellia oleifera with dynamic light adaptation adjustment, including a generation module, an analysis module, and an acquisition module. Among them, the generation module is used for preprocessing the Camellia oleifera image to be detected by region segmentation to generate a multi-region segmented Camellia oleifera image; the analysis module is used for inputting the multi-region segmented Camellia oleifera image into the multi-objective recognition immune algorithm for Camellia oleifera fruits, and the pre-trained preference multi-objective module performs light intensity analysis on the multi-region segmented Camellia oleifera image; the preference multi-objective module is used for performing target recognition on the morphological color values of Camellia oleifera fruits under different illuminations as preference data; the acquisition module is used for obtaining the recognition result of Camellia oleifera fruits in the Camellia oleifera image to be detected according to the light intensity analysis result output by the pre-trained preference multi-objective module.

[0062] Further, a multi-objective intelligent detection system for Camellia oleifera with dynamic light adaptation adjustment provided in this embodiment further includes a collection module and an extraction module. Among them, the collection module is used for collecting the dynamic light intensity data of Camellia oleifera fruits under different illuminations, and different illuminations include the light intensities in different weathers and different time periods; the extraction module is used for extracting the morphological color values corresponding to the dynamic light intensity data and inputting the morphological color values as preference data into the preference multi-objective module to be trained 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.

[0063] Further, a multi-objective intelligent detection system for Camellia oleifera with dynamic light adaptation adjustment provided in this embodiment, the generation module includes a segmentation unit and a judgment unit. Among them, the segmentation unit is used for preprocessing the Camellia oleifera image to be detected by region segmentation, dividing the Camellia oleifera image to be detected into several regions, and each region corresponds to a matching preference level; the judgment unit is used for judging whether the size of the remaining image after segmentation of the Camellia oleifera image to be detected meets the preset segmentation threshold. If the size of the remaining image after segmentation of the Camellia oleifera image to be detected is less than the preset segmentation threshold, the remaining image after segmentation of the Camellia oleifera image to be detected is incorporated into the adjacent region.

[0064] Further, in a multi-target intelligent detection system for oil-tea camellia with dynamic light-adaptive adjustment provided in this embodiment, in the analysis module, according to the immune affinity formula, the affinity clustering results of oil-tea camellia fruit targets in each region are synthesized, the oil-tea camellia fruit targets are identified from the data and the coordinate values are returned. The immune affinity formula is: (8) In formula (8), is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set as 6 preference characteristic parameters of the input R component, G component, B component, elongation, roundness and fullness of the oil-tea camellia fruit; is the oil-tea camellia image to be detected.

[0065] Further, in a multi-target intelligent detection system for oil-tea camellia with dynamic light-adaptive adjustment provided in this embodiment, in the acquisition module, based on the light intensity analysis result, selection, crossover and mutation operations are performed on the oil-tea camellia antibody population to obtain a new oil-tea camellia population, and the cell affinity is calculated in the new oil-tea camellia population. The cell affinity is: (9) (10) In formulas (9) to (10), is the new cell formed after the cell mutates, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is the exponentially decaying variable in the adjustment function, is the cell adaptive value after being normalized, is the self-adjusting mutation coefficient.

[0066] For the multi-target intelligent detection method and system for oil-tea camellia with dynamic light-adaptive adjustment provided in this embodiment, compared with the prior art, the oil-tea camellia image to be detected is preprocessed by region segmentation to generate a multi-region segmented oil-tea camellia image; the multi-region segmented oil-tea camellia image is input into the multi-target recognition immune algorithm for oil-tea camellia fruits, and the trained preference multi-target module performs light intensity analysis on the multi-region segmented oil-tea camellia image; among them, the preference multi-target module is used to perform target recognition on the morphological color values of oil-tea camellia fruits under different illuminations as preference data; according to the light intensity analysis result output by the trained preference multi-target module, the recognition result of the oil-tea camellia fruits in the oil-tea camellia image to be detected is obtained. The beneficial effects achieved by the multi-target intelligent detection method and system for oil-tea camellia with dynamic light-adaptive adjustment provided in this embodiment are as follows: 1) The recognition rate is improved by optimizing with multi-objective preference data, where the preference data corresponds to six features of the morphology and color of oil-tea camellia fruits under different light intensities, solving the problem of the dynamic change of the target recognition rate under uneven light intensity.

[0067] 2) By improving and optimizing immune algorithms such as preference level and image partitioning, the efficiency of the immune algorithm is improved. Using the efficient algorithm structure of the immune algorithm, it is applicable to the natural picking environment of oil-tea camellia with high real-time requirements and low-cost software and hardware systems.

[0068] 3) By comparing with five algorithms: 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 reaches up to 93.8% under natural light, and the recognition time is at the 160ms level, proving the usability of the algorithm in dynamic light.

[0069] The experiments and tests of the research model in this embodiment are based on more than 4,500 test images under different light environment conditions, mainly focusing on the recognition of the preference features of the color and morphological characteristics of oil-tea camellia fruits. It ensures both efficiency and speed in dynamic light recognition, providing an effective solution for the recognition of oil-tea camellia fruits in natural environments.

[0070] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A multi-objective intelligent detection method for oil tea with dynamic light adaptation adjustment, characterized in that, Including the following steps: Preprocess the oil-tea camellia image to be detected by region segmentation to generate a multi-region segmented oil-tea camellia image; Input the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits, and the trained preference multi-object module performs light intensity analysis on the multi-region segmented oil-tea camellia image; wherein, the preference multi-object module is used to perform target recognition on the morphological color values of oil-tea camellia fruits under different illuminations as preference data; According to the light intensity analysis result output by the trained preference multi-object module, obtain the recognition result of the oil-tea camellia fruits in the oil-tea camellia image to be detected.

2. The dynamic light intensity adaptive adjustment-based multi-object intelligent detection method for oil-tea camellias according to claim 1, wherein, Before the step of inputting the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits and the trained preference multi-object module performing light intensity analysis on the multi-region segmented oil-tea camellia image, it includes: Collect the dynamic light intensity data of oil-tea camellia fruits under different illuminations, and the different illuminations include the light intensities in different weathers and different time periods; Extract the morphological color values corresponding to the dynamic light intensity data, and input the morphological color values as preference data into the preference multi-object module to be trained for affinity clustering analysis training, and the morphological color values include RGB color values and morphological values; the preference data includes preference recognition data and preference levels.

3. The dynamic light-adaptive adjustment multi-object intelligent detection method for oil-tea camellia according to claim 1, characterized in that The step of preprocessing the oil-tea camellia image to be detected by region segmentation to generate a multi-region segmented oil-tea camellia image includes: Preprocess the oil-tea camellia image to be detected by region segmentation, divide the oil-tea camellia image to be detected into several regions, and each region corresponds to a matching preference level; Judge whether the size of the remaining image after the oil-tea camellia image to be detected is segmented meets the preset segmentation threshold. If the size of the remaining image after the oil-tea camellia image to be detected is segmented is smaller than the preset segmentation threshold, incorporate the remaining image after the oil-tea camellia image to be detected into the adjacent region.

4. The multi-objective intelligent detection method for oil tea with dynamic light-adaptive adjustment according to claim 1, wherein In the step of inputting the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits and the trained preference multi-object module performing light intensity analysis on the multi-region segmented oil-tea camellia image, according to the immune affinity formula, synthesize the affinity clustering results of the oil-tea camellia fruit targets in each region, identify the oil-tea camellia fruit targets from the data and return the coordinate values, and the immune affinity formula is: ； Among them, is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set as the 6 preference characteristic parameters of the input R component, G component, B component, elongation, roundness and perfection of the oil-tea fruit; is the oil-tea image to be detected.

5. The dynamic light-adaptive adjustment-based multi-object intelligent detection method for oil tea as claimed in claim 1, wherein, In the step of obtaining the recognition result of the oil-tea camellia fruits in the oil-tea camellia image to be detected according to the light intensity analysis result output by the trained preference multi-object module, based on the light intensity analysis result, perform selection, crossover and mutation operations on the oil-tea camellia antibody population to obtain a new oil-tea camellia population, and calculate the cell affinity in the new oil-tea camellia population, and the cell affinity is: ； Among them, is a new cell formed after cell mutation, is the original cell, is the exponentially decaying variable in the adjustment function, is the cell adaptation value after standardization, is the self-regulating mutation coefficient.

6. An intelligent multi-object detection system for oil tea with dynamic light-adaptive adjustment, characterized in that, Including: A generation module for preprocessing the oil-tea camellia image to be detected by region segmentation to generate a multi-region segmented oil-tea camellia image; An analysis module for inputting the multi-region segmented oil-tea camellia image into the multi-object recognition immune algorithm for oil-tea camellia fruits, and the trained preference multi-object module performs light intensity analysis on the multi-region segmented oil-tea camellia image; the preference multi-object module is used to perform target recognition on the morphological color values of oil-tea camellia fruits under different illuminations as preference data; An acquisition module, configured to obtain the recognition result of the oil-tea fruit in the to-be-detected oil-tea image according to the light intensity analysis result output by the trained preference multi-object module.

7. The dynamic light-adaptive adjustment multi-object intelligent detection system for oil-tea camellia according to claim 6, characterized in that The oil-tea multi-object intelligent detection system with dynamic light adaptive adjustment further includes: A collection module, configured to collect the dynamic light intensity data of the oil-tea fruit under different light conditions, where the different light conditions include the light intensities in different weather and different time periods; An extraction module, configured to extract the morphological color values corresponding to the dynamic light intensity data, and use the morphological color values as preference data to input into the to-be-trained preference multi-object module for affinity clustering analysis training, where the morphological color values include RGB color values and morphological values; the preference data includes preference recognition data and preference levels.

8. The dynamic light intensity adaptive adjustment multi-object intelligent detection system for oil-tea camellia according to claim 6, characterized in that, The generation module includes: A segmentation unit, configured to perform regional segmentation preprocessing on the to-be-detected oil-tea image, and segment the to-be-detected oil-tea image into several regions, with each region corresponding to a matching preference level; A judgment unit, configured to judge whether the size of the remaining image after segmentation of the to-be-detected oil-tea image meets a preset segmentation threshold. If the size of the remaining image after segmentation of the to-be-detected oil-tea image is smaller than the preset segmentation threshold, the remaining image after segmentation of the to-be-detected oil-tea image is incorporated into the adjacent region.

9. The dynamic light adaptive adjustment oil-tea camellia multi-target intelligent detection system according to claim 6, wherein In the analysis module, according to the immune affinity formula, the affinity clustering results of the oil-tea fruit targets in each region are integrated, and the oil-tea fruit targets are identified from the data and the coordinate values are returned. The immune affinity formula is: ； Among them, is the affinity value, is the Gaussian function symbol, is the Gaussian function parameter, is set as the 6 preference characteristic parameters of the input R component, G component, B component, elongation, roundness and fullness of the oil-tea fruit; is the oil-tea image to be detected.

10. The dynamic light-adaptive adjusted multi-object intelligent detection system for oil-tea camellia according to claim 6, characterized in that, In the acquisition module, based on the light intensity analysis result, selection, crossover, and mutation operations are performed on the oil-tea antibody population to obtain a new oil-tea population, and the cell affinity is calculated in the new oil-tea population. The cell affinity is: ； Among them, is a new cell formed after cell mutation, is the original cell, is a Gaussian random variable with a mean of 0 and a deviation of 1; is an exponentially decaying variable in the adjustment function, is the cell adaptation value after standardization, is the self-adjusting mutation coefficient.

Images