Intelligent annona squamosa cultivation and energy-saving collection system based on machine vision

Abstract

The invention discloses an intelligent cultivation and energy-saving collection system for annona squamosa based on machine vision. The intelligent cultivation and energy-saving collection system comprises an orchard vision module, an individual vision module, a control module, an auxiliary module, a central processing module and a communication module, wherein the orchard vision module is combined with ground fixed camera equipment and a control unmanned aerial vehicle to realize accurate positioning and three-dimensional reconstruction of the annona squamosa orchard; through a target detection algorithm, the individual vision module not only can confirm the optimal picking opportunity, but also can intelligently diagnose diseases and pests and perform intelligent prevention and control; the auxiliary module comprises a vision enhancement module and an exposure optimization processor and is used for improving the stability and accuracy of image acquisition equipment in the vision module; the influence of wind power is considered, the speed or acceleration of the annona squamosa fruits is obtained through the central calculation module by means of the wind power, the movement track of the annona squamosa fruits is predicted, the movement trend of the annona squamosa fruits is calculated in combination with different movement assumptions (uniform speed and uniform acceleration), and the grabbing time and speed of the mechanical arm are optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of sugar apple picking and machine vision, and particularly relates to a sugar apple intelligent cultivation and energy-saving collection system based on machine vision. Background Art

[0002] Automated sugar-apple harvesting technology has been a key area of ​​intelligent agricultural development in recent years. Traditionally, sugar-apple harvesting relies on manual labor, which is labor-intensive and inefficient. With the advancement of agricultural mechanization, automated harvesting devices have emerged, primarily relying on the machine's initiative to complete the harvesting task and then return to the designated location. However, these devices typically operate in an active mode, unable to operate continuously for long periods of time, and consume significant amounts of energy during operation. Therefore, increasing the operating time of automated harvesting devices, reducing energy consumption, and achieving fully automated operational decision-making remain key challenges in the development of this technology.

[0003] Existing automated sugar-apple harvesting technology still has some shortcomings when dealing with large orchards and large numbers of fruit. In particular, sugar-apple fruit is susceptible to wind, which can cause the fruit to sway, leading to unstable positions during harvesting. Existing technologies fail to fully account for the impact of wind on fruit position and lack the ability to predict and analyze wind direction and wind speed changes in real time. An ideal automated harvesting system should be able to predict the position of the fruit based on the wind's impact on the fruit and adjust the movement speed of the robotic arm accordingly, ensuring that the robotic arm closest to the device is at a specific movement speed to precisely grasp the fruit. Furthermore, existing systems still rely on manual judgment of suitability for the operation and fail to implement intelligent climate-adaptive decision-making, resulting in insufficient optimization of operating efficiency and energy consumption. Therefore, combining wind direction prediction with intelligent decision-making systems to reduce energy consumption and extend operation time remains a key issue that needs to be addressed in automated sugar-apple harvesting technology. Furthermore, during sugar-apple harvesting, traditional visual recognition systems can suffer from reduced recognition accuracy and unstable target tracking due to varying lighting conditions, fruit obstruction, and background interference. These are also issues that require attention in automated sugar-apple harvesting technology. Summary of the Invention

[0004] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology and provide a sugar apple intelligent cultivation and energy-saving collection system based on machine vision, so as to realize the automated cultivation and picking of sugar apple fruits and improve the level of automated management of orchards.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A machine vision-based sugar apple intelligent cultivation and energy-saving collection system includes an orchard vision module, an individual vision module, a control module, an auxiliary module, a central processing module, and a communication module;

[0007] The control module is connected to the orchard vision module, the individual vision module, the central processing module and the communication module respectively; the auxiliary module is connected to the orchard vision module and the individual vision module respectively;

[0008] The orchard vision module uses ground-based fixed cameras and aerial drones to accurately locate and model the sugar apple orchard using computer vision technology and three-dimensional reconstruction algorithms, thereby obtaining the three-dimensional spatial structure of the sugar apple orchard and the location and distribution of the sugar apple fruits.

[0009] The individual vision module is mounted on the automated picking equipment, and uses a first target detection algorithm to detect the maturity of the sugar apple fruit, and uses a second target detection algorithm to identify pests and diseases on the sugar apple and sugar apple leaves;

[0010] The auxiliary module includes a visual enhancement module and an exposure optimization processor; the visual enhancement module enhances the collected visual image based on the light sensing data of the orchard vision module or the individual vision module; the exposure optimization processor is used to perform intelligent exposure control on the light of the orchard vision module or the individual vision module;

[0011] The central processing module calculates the speed at which the robotic arm of the automated picking device just picks the sugar-apple fruit based on the motion trajectory of the sugar-apple fruit blown by the wind; the motion trajectory of the sugar-apple fruit blown by the wind is an ellipse or an arc on a circle;

[0012] The communication module uses low-latency 5G communication technology to be responsible for data transmission and information sharing between modules;

[0013] The control module is used to control and coordinate the work among the modules.

[0014] As a preferred technical solution, the ground fixed camera device is a high-resolution camera or a laser radar, which is used to collect ground data of the sugar apple orchard; the ground data is image and depth data of a local area of ​​the sugar apple orchard;

[0015] The aerial drone captures aerial data of the sugar apple orchard from the air and generates three-dimensional point cloud data through structured light technology or stereo vision; the aerial data is a high-definition image of the sugar apple orchard;

[0016] The orchard vision module integrates ground data and aerial data, and applies SLAM technology and deep learning algorithms to reconstruct the three-dimensional spatial structure of the sugar apple orchard and identify the location and distribution of sugar apple fruits.

[0017] As a preferred technical solution, the visual enhancement module enhances the collected visual image based on the light sensing data of the orchard visual module or the individual visual module, specifically:

[0018] An environmental quality evaluation model is constructed based on light sensor data and image feature analysis, and the environmental state function is defined:

[0019]

[0020] Where Φ is the environmental quality evaluation model, L is the representation value of light intensity after nonlinear mapping, C is the fog concentration index based on dark channel statistics, N is the noise assessment value combined with frequency domain analysis and light sensor readings, γ is the noise sensitivity coefficient corresponding to the noise assessment value N in the environmental quality score, α(t) and β(t) are dynamic weight coefficients, which are adaptively adjusted according to the circadian rhythm to meet the following requirements:

[0021]

[0022] k1 is the light weight adjustment rate coefficient, L target The target light level set for the model, L current The real-time light level measurement value of the light sensor;

[0023] Construct a decision function and use the dark light enhancement branch to perform dark light enhancement on the collected visual image or optimize the transmittance for haze interference based on the overall image brightness and the real-time light level measurement value of the light sensor to obtain a basic enhanced image;

[0024] A gradient domain blending strategy is used to fuse high-frequency details with basic enhancement results to obtain a fused enhanced image.

[0025] A reflection suppression model is constructed to suppress reflections on the fused enhanced image to obtain a visually enhanced image.

[0026] As a preferred technical solution, when the overall brightness of the visual image is lower than the set brightness threshold, and the real-time light level measurement value of the light sensor is lower than the set light level threshold, the environment of the visual image is judged to be a dark light scene. When the environmental state function Q is less than the lower threshold value θ of the environmental state function Q, the light level of the visual image is judged to be a dark light scene. low When , the decision function uses the dark light enhancement branch to perform dark light enhancement on the collected visual image, specifically:

[0027] A dual-channel fusion model is introduced to enhance dark light of the collected visual images:

[0028] E(x,y)=ω·T cnn (I)+(1-ω)·R msr (I),

[0029] Among them, (x, y) is the coordinate of a pixel in the captured image, T cnn is the enhanced result of the lightweight convolutional network output, R msr To improve the output of the multi-scale Retinex algorithm; ω is the fusion weight, which is dynamically calculated by the local contrast σ(x,y) of the image:

[0030]

[0031] k2 is the sensitivity coefficient that controls the speed at which the fusion weight ω changes with the local contrast, and σ0 is the set local contrast threshold.

[0032] As a preferred technical solution, when the overall brightness of the visual image is within the normal brightness range but there are obvious low-frequency blur features, and the dark channel map statistics show that the fog concentration is higher than the set fog concentration threshold, the environment of the visual image is judged to be a fog and haze interference environment. The decision function optimizes the transmittance for fog and haze interference, specifically:

[0033] Define the corrected transmittance estimation function:

[0034] t′(x)=max(t(x),η·G(x)⊙M fruit ),

[0035] Among them, G(x) is the regional growth function of the sugar apple fruit, M fruit is the probability map of the sugar apple fruit image output by the mask generator, η is the canopy penetration compensation factor, and ⊙ is the Hadamard product;

[0036] Use the corrected transmittance estimation function to restore the visual image and obtain a restored clear image;

[0037] The texture preservation constraint term is introduced to optimize the modified transmittance estimation function:

[0038]

[0039] Where I is the visual image, J is the estimated value of the restored clear image, t′ is the corrected transmittance estimation function, A() is the estimated value of the global background light, λ is the weight coefficient of the texture preservation constraint term, S texture is the image texture guide map.

[0040] As a preferred technical solution, the high-frequency details are extracted by applying high-pass filtering or edge detection to the visual image; the fusion process is expressed as:

[0041]

[0042] Among them, J finalis the fusion enhanced image, μ is the adjustment coefficient of the fusion enhancement strength, sign(D) is used to retain the change direction of the high-frequency details D. If the sign() input is a positive number, it returns +1; if the sign() input is a negative number, it returns -1; if the sign() input is zero, it returns 0;

[0043] The reflection suppression model is expressed as:

[0044]

[0045] Where R suppress is the reflection suppression model, is the brightness channel of the fused enhanced image in the HSV color space, ρ is the peel reflection threshold of the sugar apple fruit, and n is the nonlinear suppression strength coefficient.

[0046] As a preferred technical solution, the exposure optimization processor is used to perform intelligent exposure control on the illumination of the orchard vision module or the individual vision module, specifically:

[0047] Based on the spectral characteristics of photosynthetically active radiation, an exposure quality evaluation model is constructed, and the pixel-level exposure anomaly index is defined:

[0048]

[0049] Where Q exp (x) is the pixel-level exposure anomaly index at pixel x in the visual image collected by the orchard vision module or the individual vision module, L(x) is the light intensity value of pixel x, V(x) is the chlorophyll fluorescence intensity value at pixel x, σ v is the standard deviation of chlorophyll fluorescence intensity, V opt is the optimal value of chlorophyll fluorescence intensity, μ PAR is the expected value of photosynthetically active radiation, which is dynamically updated by the chlorophyll fluorescence dynamics model:

[0050]

[0051] in, is the biomass change rate of the plant, F(t) is the chlorophyll fluorescence intensity at time t, and F max is the maximum fluorescence intensity value under light saturation state, k p is the rate coefficient of photosynthetic active radiation renewal;

[0052] Design a dual-channel nonlinear transfer function to process overexposed and underexposed areas separately:

[0053]

[0054] Among them, T(I) is the enhanced result of the output visual image, I γ(t)is the image after dynamic gamma correction, γ(t) is the dynamic gamma correction parameter, β is the sensitivity coefficient of brightness adjustment, I is the brightness value of the input visual image, I th is the threshold value of overexposure and underexposure areas, and the dynamic adjustment parameters γ(t) and α(t) are constrained by photosynthetic efficiency:

[0055]

[0056] is the plant biomass change rate, is the sensitivity of chlorophyll fluorescence intensity to light intensity, k f is the sensitivity coefficient of photosynthesis to light intensity, F0 is the initial fluorescence intensity of photosynthesis;

[0057] In response to the need for continuous data collection in agricultural monitoring, a time series smoothing constraint based on metabolic rhythm is proposed:

[0058]

[0059] Among them, I t is the image at time t, I t-1 is the image at time t-1, ||·|| W is the weighted Euclidean distance, is the gradient of the pixel-level exposure anomaly index at time t;

[0060] By fusing multispectral sensor data with RGB image features, the light intensity distribution in the photosynthetically active radiation band is reconstructed:

[0061]

[0062] Among them, I PAR is the light intensity distribution in the photosynthetically active radiation band, λ is the photosynthetically active radiation band, PAR λ is the radiation intensity in the photosynthetically active radiation band, I RGB is the image collected by the orchard vision module or the individual vision module, ξ is the scaling factor, and the weight function w(λ) is determined by the gradient derived from the chlorophyll absorption spectrum φ(λ):

[0063]

[0064] is the derivative of the weight function with respect to the photosynthetically active radiation band, k w is the weight coefficient, is the gradient of the chlorophyll absorption spectrum, is the rate of change of plant growth rate over time; CNN spec It is a lightweight spectral prediction network used to estimate PAR components from RGB images when there is no multispectral sensor.

[0065] As a preferred technical solution, the speed at which the robotic arm of the automated picking equipment just picks the sugar-apple fruit is calculated based on the motion trajectory of the sugar-apple fruit blown by the wind, specifically:

[0066] Let the center of the ellipse or circle corresponding to the arc trajectory formed by the wind of the sugar apple fruit be point P, let the critical points of the arc trajectory on the ellipse or circle be points A and B respectively, let the coordinates of the position of the automated picking equipment be Q, then the intersection of the straight line PQ and the partial circle is point R, let the moving speed of the robotic arm of the automated picking equipment be v L constant;

[0067] When the sugar apple fruit is blown by the wind on an ellipse or circle to perform uniform circular motion and uniform accelerated circular motion, the moving speed of the robotic arm is calculated based on the condition that the time it takes for the sugar apple fruit to move to point R is the same as the time it takes for the robotic arm to move to point R.

[0068] As a preferred technical solution, the calculation to obtain the moving speed of the robotic arm is specifically as follows:

[0069] When the sugar apple fruit is blown by the wind and performs uniform circular motion on the circle, let the angular velocity of the sugar apple fruit be ω ab Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0070] Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R.

[0071] Determine the movement process of the sugar apple fruit on the circular trajectory. Assume that the process of the sugar apple fruit moving from point A to point B or from point B to point A on the circular trajectory is one movement of the sugar apple fruit on the circular trajectory. If the sugar apple fruit and the robotic arm arrive at point R at the same time when the sugar apple fruit moves on the circular trajectory for the nth time, when n = 2k-1, the sugar apple fruit just moves from point A to point R. Calculate the movement time of the sugar apple fruit. When n=2k, the sugar apple fruit just moves from point B to point R. Calculate the movement time of the sugar apple fruit. Where n and k are positive numbers, and k is not equal to 0;

[0072] The motion time of the sugar apple fruit is expressed as:

[0073]

[0074] The moving speed v of the robotic arm is calculated based on the movement time t1 of the sugar apple fruit L ;

[0075] When the sugar apple fruit is blown by the wind on a circle to perform uniform accelerated circular motion, assuming that the angular acceleration α of the sugar apple fruit is constant, according to the speed v of the sugar apple fruit at point A and point B, A ,v B Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0076] Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R.

[0077] Similarly, the motion process of the sugar apple fruit on the arc trajectory is determined to obtain the motion time t2 of the sugar apple fruit as:

[0078]

[0079] The moving speed v of the robotic arm is calculated based on the movement time t2 of the sugar apple fruit L ;

[0080] When the sugar apple fruit is blown by the wind and performs uniform circular motion on the ellipse, let the angular velocity of the sugar apple fruit be ω ab Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0081] Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move to point R.

[0082] Determine the motion process of the sugar apple fruit on the arc trajectory. Similarly, the motion time t3 of the sugar apple fruit is obtained as:

[0083]

[0084] The moving speed v of the robotic arm is calculated based on the movement time t3 of the sugar apple fruit L ;

[0085] When the sugar apple fruit is blown by the wind and performs uniformly accelerated circular motion on the ellipse, the tangential velocity a of the sugar apple fruit is obtained. t Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0086] Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R.

[0087] Similarly, the movement process of the sugar apple fruit on the arc trajectory is judged, and the movement time t4 of the sugar apple fruit is obtained as:

[0088]

[0089] The moving speed v of the robotic arm is calculated based on the movement time t4 of the sugar apple fruit L .

[0090] As a preferred technical solution, the individual vision module is composed of a camera and an intelligent chip;

[0091] The pests and diseases include anthracnose, blank canker, beam rot, fruit leaf spot, leaf spot and whitefly;

[0092] The automated picking equipment is equipped with a medicine spraying facility; when the second target detection algorithm of the individual visual module identifies pests and diseases on the sugar apple and sugar apple leaves, the corresponding medicine is sprayed to treat and eliminate the pests.

[0093] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0094] 1. Accurate orchard modeling based on multi-source data fusion

[0095] The orchard vision module combines fixed ground cameras with controlled drones to achieve precise positioning and 3D reconstruction of the sugar-apple orchard. The fixed ground cameras use high-resolution cameras or laser radar (LiDAR) to collect ground data such as local images and depth information, while aerial drones capture aerial data such as large-scale high-definition images of the sugar-apple orchard, generating large-scale 3D point cloud data through structured light technology or stereo vision technology. Finally, the orchard vision module uses SLAM (Simultaneous Localization and Mapping) technology combined with deep learning algorithms to achieve real-time 3D reconstruction of the orchard environment, improving the detection accuracy and spatial distribution analysis capabilities of sugar-apple fruits, providing high-precision navigation and path planning basis for automated picking equipment, and improving the accuracy of fruit positioning.

[0096] 2. Target detection and pest identification

[0097] The individual vision module of this application uses two sets of target detection algorithms, which can not only judge the maturity of the fruit to confirm the best time for picking and reduce the error of human judgment, but also intelligently diagnose pests and diseases and carry out intelligent prevention and control. It can accurately spray corresponding potions for different pests and diseases, realize the automated cultivation of sugar apples, and improve the level of automation in orchard management.

[0098] 3. Visual enhancement and intelligent exposure

[0099] The auxiliary modules in this application include a visual enhancement module and an exposure optimization processor, which are used to improve the stability and accuracy of the image acquisition equipment in the orchard vision module and the individual vision module, ensuring high-precision target detection capabilities in complex lighting environments. Among them, the visual enhancement module adopts an environmental perception-dynamic decision-making-hybrid enhancement architecture, combined with contrast enhancement, haze removal, reflection suppression and other technologies to improve the visibility of fruits in complex environments; and the exposure optimization processor (PEOM) dynamically adjusts the exposure parameters based on the plant photosynthetic characteristics and photosynthetically active radiation (PAR) model to ensure stable imaging quality of fruits under different lighting conditions, and can maintain high-precision target detection capabilities even when the lighting changes drastically, the background is complex or the occlusion is severe.

[0100] 4. Picking trajectory of the robotic arm under the influence of wind

[0101] This application takes into account the influence of wind power, uses wind power through a central computing module to obtain the speed or acceleration of the sugar apple fruit, and predicts its motion trajectory. Combining different motion assumptions (uniform speed, uniform acceleration), the movement trend of the sugar apple fruit is calculated, and the grasping timing and speed of the robotic arm are optimized; by constructing an optimal gripping model, the minimum starting speed of the robotic arm is calculated to ensure that the robotic arm just grasps the fruit at the point closest to the sugar apple fruit, while reducing the ineffective movement of the robotic arm to save energy, and improving the working efficiency so that the robotic arm can dynamically adjust the grasping speed to ensure efficient picking. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0103] Figure 1 The figure is a structural diagram of the sugar apple intelligent cultivation and energy-saving collection system based on machine vision in an embodiment of the present invention.

[0104] Figure 2 Schematic diagram of the movement of the sugar apple with a circular motion trajectory in an embodiment of the present invention.

[0105] Figure 3 Schematic diagram of the movement of the sugar apple with an elliptical motion trajectory in an embodiment of the present invention. DETAILED DESCRIPTION

[0106] In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0107] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0108] like Figure 1 As shown, this embodiment provides a sugar apple intelligent cultivation and energy-saving collection system based on machine vision, including an orchard vision module, an individual vision module, a control module, an auxiliary module, a central processing module and a communication module; wherein the control module is respectively connected to the orchard vision module, the individual vision module, the central processing module and the communication module; the auxiliary module is respectively connected to the orchard vision module and the individual vision module.

[0109] For the orchard vision module, it uses computer vision technology and three-dimensional reconstruction algorithms through ground-fixed camera equipment and aerial drones to accurately locate and model the sugar apple orchard, obtain the three-dimensional spatial structure of the sugar apple orchard and the location and distribution of the sugar apple fruits.

[0110] Furthermore, the ground-fixed camera equipment is a high-resolution camera or a laser radar (LiDAR), which is used to collect ground data of the sugar-apple orchard; the ground data is the image and depth data of the local area of ​​the sugar-apple orchard; and the aerial drone shoots the aerial data of the sugar-apple orchard from the air, and generates three-dimensional point cloud data through structured light technology or stereo vision; the aerial data is a large-scale high-definition image of the sugar-apple orchard; based on the ground data and control data, the orchard vision module fuses the two data and applies SLAM (simultaneous localization and mapping) technology and deep learning algorithms to reconstruct the three-dimensional spatial structure of the sugar-apple orchard, identify the position and distribution of the sugar-apple fruits, and provide a data basis for the subsequent path planning of the automated picking equipment.

[0111] As for the individual vision module, it is installed on the automated picking equipment and uses the first target detection algorithm to detect the maturity of the sugar apple fruit, and uses the second target detection algorithm to identify diseases and pests of the sugar apple and sugar apple leaves.

[0112] Furthermore, the individual vision module consists of a camera and an intelligent chip, and uses two sets of target detection algorithms to achieve two different detection tasks; the identified pests and diseases include anthrax, white canker, beam leaf rot, fruit leaf spot, leaf spot, and whitefly. In this embodiment, the first target detection algorithm and the second target detection algorithm can be target detection networks such as yolov3 and yolov12. Of course, other target detection networks can also be used to achieve the functions of this module, and this application will not go into details.

[0113] Furthermore, the automated harvesting equipment is equipped with a spraying system. When the second target detection algorithm of the individual vision module identifies pests and diseases on the sugar apples and sugar apple leaves, the spraying system is controlled to spray the corresponding pesticide to treat the pests. For example, carbendazim or thiophanate-methyl are used for anthracnose; Bordeaux mixture or thiophanate-methyl are used for whitefly canker; thiophanate-methyl or thiophanate-methyl are used for beam rot; chlorothalonil or thiophanate-methyl are used for fruit leaf spot; thiophanate-methyl or thiophanate-methyl are used for leaf spot; and imidacloprid or chlorantraniliprole are used for whiteflies.

[0114] As for the auxiliary module, it includes a visual enhancement module and an exposure optimization processor; wherein, the visual enhancement module enhances the collected visual image based on the light sensing data of the orchard vision module or the individual vision module; the exposure optimization processor is used to perform intelligent exposure control on the light of the orchard vision module or the individual vision module.

[0115] Furthermore, the visual enhancement module achieves visual enhancement in complex environments through a multi-algorithm collaborative mechanism, establishing a three-level processing architecture of environmental perception-dynamic decision-making-hybrid enhancement. Specifically:

[0116] First, an environmental quality evaluation model is constructed based on light sensor data and image feature analysis, and the environmental state function is defined:

[0117]

[0118] Where Φ is the environmental quality evaluation model, L is the representation value of light intensity after nonlinear mapping, C is the fog concentration index based on dark channel statistics, N is the noise evaluation value combined with frequency domain analysis and light sensor readings, γ is the noise sensitivity coefficient corresponding to the noise evaluation value N in the environmental quality score, α(t) and β(t) are dynamic weight coefficients, which are adaptively adjusted according to the circadian rhythm to meet the following requirements:

[0119]

[0120] k1 is the light weight adjustment rate coefficient, L target The target light level set for the model, L currentThe real-time light level measurement value of the light sensor.

[0121] To improve visual perception quality, we need to select an appropriate enhancement strategy based on the image characteristics. Therefore, we construct a decision function. Based on the overall image brightness and the real-time light level measurements from the illumination sensor, we use the dark light enhancement branch to perform dark light enhancement on the captured visual image or optimize transmittance for haze interference, resulting in a basic enhanced image. The dark light enhancement branch is primarily used in low-light, dark environments, while transmittance optimization is used to address haze or other visually blurred interference scenarios.

[0122] Furthermore, for the dark light enhancement branch, when the overall brightness of the visual image is lower than the set brightness threshold and the real-time light level measurement value of the light sensor is lower than the set light level threshold, the environment of the visual image is judged to be a dark light scene. When the environmental state function Q is less than the lower threshold θ of the environmental state function Q, the dark light scene is judged to be a dark light scene. low Then the decision function uses the dark light enhancement branch to enhance the dark light of the collected visual image, improve the brightness and restore the details, specifically:

[0123] A dual-channel fusion model is introduced to enhance dark light of the collected visual images:

[0124] E(x,y)=ω·T cnn (I)+(1-ω)·R msr (I),

[0125] Among them, (x, y) is the coordinate of a pixel in the captured image, T cnn is the enhanced result of the lightweight convolutional network output, R msr To improve the output of the multi-scale Retinex algorithm; ω is the fusion weight, which is dynamically calculated by the local contrast σ(x,y) of the image:

[0126]

[0127] k2 is the sensitivity coefficient that controls the speed at which the fusion weight ω changes with the local contrast, and σ0 is the set local contrast threshold.

[0128] Furthermore, when the overall brightness of the visual image is within the normal brightness range but there are obvious low-frequency blur features, and the dark channel map statistics show that the fog concentration is higher than the set fog concentration threshold, the visual image environment is judged to be a haze interference environment. The decision function proposes a transmittance optimization model based on agricultural scene priors to enhance vision in response to haze interference. Specifically:

[0129] Define the corrected transmittance estimation function:

[0130] t'(x)=max(t(x),η·G(x)⊙M fruit),

[0131] Among them, G(x) is the regional growth function of the sugar apple fruit, M fruit is the probability map output by the mask generator of the sugar apple fruit image, η is the canopy penetration compensation factor, and ⊙ is the Hadamard product. In this embodiment, the sugar apple fruit region growth function G(x) is obtained by combining the growth characteristics of sugar apple with image analysis technology: first, sugar apple fruit image data at different growth stages are collected, and the fruit region is identified using image segmentation and feature extraction technology; then, a mathematical model describing the fruit growth process is constructed using experimental data or based on a plant growth model (such as a logistic model); finally, combined with machine learning or optimization algorithms, the growth function G(x) is dynamically updated according to the collected images and environmental data to improve its accurate description of the fruit region and the transmittance optimization effect.

[0132] Use the modified transmittance estimation function to restore the visual image and obtain a restored clear image;

[0133] The texture preservation constraint term is introduced to optimize the modified transmittance estimation function:

[0134]

[0135] Where I is the visual image, J is the estimated value of the restored clear image, t′ is the corrected transmittance estimation function, A() is the estimated value of the global background light (atmospheric light), which is used to compensate for the influence of light scattering; λ is the weight coefficient of the texture preservation constraint term, S texture is an image texture guide map, which is used to maintain the detail structure in the transmittance optimization. In this embodiment, the image texture guide map S is obtained. texture The method is to perform texture analysis and feature extraction on the visual image I: first, the texture features of the image are extracted by edge detection texture analysis or wavelet transform to obtain the local texture information of the image; then, the area with obvious texture features in the image is extracted by image segmentation or region growing technology, and it is converted into a texture guidance map S. texture , which is used to guide the texture-preserving optimization process of the modified transmittance estimation function.

[0136] After obtaining the basic enhanced image, the gradient domain blending strategy is used to fuse the high-frequency details D with the basic enhanced result J to obtain a fused enhanced image. The process is expressed as:

[0137]

[0138] Among them, J finalis the fusion-enhanced image, μ is the adjustment coefficient for the fusion-enhancement strength, and sign(D) is used to preserve the direction of change in the high-frequency detail D. If the sign() input is positive, it returns +1; if the sign() input is negative, it returns -1; if the sign() input is zero, it returns 0. The high-frequency detail D is extracted by applying high-pass filtering or edge detection to the input visual image, primarily used to preserve texture and edge information. The basic enhancement result J is the image processed through transmittance optimization or dark light enhancement, which is used to improve image brightness and clarity.

[0139] In order to adapt to the optical properties of the sugar apple surface, a reflection suppression model is finally constructed to suppress reflections on the fused enhanced image to obtain a visually enhanced image. The reflection suppression model is expressed as:

[0140]

[0141] Where R suppress is the reflection suppression model, is the brightness channel of the fused enhanced image in the HSV color space, ρ is the peel reflection threshold of the sugar apple fruit, and n is the nonlinear suppression strength coefficient.

[0142] Therefore, the visual enhancement module forms a multi-scale, multi-physical quantity collaborative enhancement mechanism through the above model, which jointly ensures that in complex agricultural and forestry scenarios, the visual enhancement processing results meet both the needs of machine vision analysis and the fidelity requirements of crop physiological characteristics.

[0143] Furthermore, the exposure optimization processor (PEOM) solves the problem of exposure anomalies caused by complex lighting conditions in agricultural scenes by establishing a deep coupling mechanism between plant physiological characteristics and optical imaging. Its core lies in: converting photosynthetic response characteristics into mathematical constraints to achieve intelligent exposure control that conforms to the laws of plant growth. This application realizes adaptive exposure control by combining exposure anomaly index, exponential smoothing term, temporal smoothing constraint and transfer function; when the health status of the plant (through indicators such as chlorophyll fluorescence intensity) changes, the exposure optimization processor can dynamically adjust the exposure parameters of the image capture to ensure that the exposure of each pixel is within a reasonable range, thereby optimizing the visual acquisition effect of the orchard. Specifically including:

[0144] 1. Assessment of Photosynthetic Response to Light Exposure

[0145] Based on the spectral characteristics of photosynthetically active radiation (PAR) (400-700nm), an exposure quality evaluation model is constructed, and the pixel-level exposure anomaly index is defined:

[0146]

[0147] Where Q exp(x) is the pixel-level exposure anomaly index at pixel x in the visual image collected by the orchard vision module or the individual vision module, L(x) is the light intensity value of pixel x, V(x) is the chlorophyll fluorescence intensity value at pixel x, σ v is the standard deviation of chlorophyll fluorescence intensity, V opt is the optimal value of chlorophyll fluorescence intensity, μ PAR is the expected value of photosynthetically active radiation, which is dynamically updated by the chlorophyll fluorescence dynamics model:

[0148]

[0149] in, is the biomass change rate of the plant, F(t) is the chlorophyll fluorescence intensity at time t, and F max is the maximum fluorescence intensity value under light saturation state, k p The rate coefficient for updating photosynthetically active radiation; the exposure quality evaluation model dynamically adjusts the optimal brightness range by monitoring plant photosynthetic activity to ensure that exposure correction always serves the needs of crop growth monitoring.

[0150] 2. Metabolic Constrained Exposure Correction Algorithm

[0151] Design a dual-channel nonlinear transfer function to process overexposed and underexposed areas separately:

[0152]

[0153] Among them, T(I) is the enhanced result of the output visual image, I γ(t) is the image after dynamic gamma correction, γ(t) is the dynamic gamma correction parameter, β is the sensitivity coefficient of brightness adjustment, I is the brightness value of the input visual image, I th is the threshold value of overexposure and underexposure areas, and the dynamic adjustment parameters γ(t) and α(t) are constrained by photosynthetic efficiency:

[0154]

[0155] is the plant biomass change rate, is the sensitivity of chlorophyll fluorescence intensity to light intensity, k f is the sensitivity coefficient of photosynthesis to light intensity, and F0 is the initial fluorescence intensity of photosynthesis. This design automatically enhances overexposure inhibition in the light inhibition stage (when the light intensity exceeds the tolerance of the photosynthetic system), and prioritizes compensation for underexposed areas in the light limitation stage (when the light intensity is insufficient), thereby achieving exposure optimization that is consistent with the physiological state of the plant.

[0156] 3. Temporal consistency and spectral fidelity mechanism

[0157] In response to the need for continuous data collection in agricultural monitoring, a time series smoothing constraint based on metabolic rhythm is proposed:

[0158]

[0159] Among them, I t is the image at time t, I t-1 is the image at time t-1, ||·|| W is the weighted Euclidean distance, is the gradient of the pixel-level exposure anomaly index at time t. This constraint allows for larger exposure adjustments during rapid growth phases (where metabolic changes are significant), while maintaining strict consistency during stable growth phases to avoid interference with growth trend analysis due to exposure jumps.

[0160] 4. Cross-modal reconstruction of photosynthetically active radiation

[0161] By fusing multispectral sensor data with RGB image features, the light intensity distribution in the photosynthetically active radiation band is reconstructed:

[0162]

[0163] Among them, I PAR is the light intensity distribution in the photosynthetically active radiation band, λ is the photosynthetically active radiation band, PAR λ is the radiation intensity in the photosynthetically active radiation band, I RGB is the image collected by the orchard vision module or the individual vision module, ξ is the scaling factor used to adjust the matching degree between the spectral information output by the CNN network (PAR component estimated from the RGB image) and the actual multispectral data, and the weight function w(λ) is determined by the gradient derived from the chlorophyll absorption spectrum φ(λ):

[0164]

[0165] is the derivative of the weight function with respect to the photosynthetically active radiation band, indicating the change of weight when the spectral wavelength changes; k w is the weight coefficient, which is used to control the intensity of weight change; is the gradient of the chlorophyll absorption spectrum, which indicates the change in the light absorption rate of plant chlorophyll to different photosynthetically active radiation bands. is the rate of change of plant growth rate over time, indicating the changes in photosynthesis and light absorption during plant growth; CNN spec It is a lightweight spectral prediction network used to estimate PAR components from RGB images when there is no multispectral sensor, ensuring the applicability of the system on low-cost hardware.

[0166] For the central processing module, it calculates the speed at which the robotic arm of the automated picking equipment just picks the sugar-apple fruit based on the movement trajectory of the sugar-apple fruit blown by the wind; this application sets the movement trajectory of the sugar-apple fruit blown by the wind as an ellipse or an arc on a circle.

[0167] Furthermore, considering that the trajectory of the sugar apple fruit due to wind movement is an ellipse or an arc on a circle, this application provides the control speed of the robotic arm under different assumptions. Starting from the minimum movement speed achieved by the robotic arm, when the device meets the sugar apple fruit on the arc of the sugar apple fruit movement, the sugar apple fruit is just grasped. The calculation steps are:

[0168] a. Figure 2 ,like Figure 3 As shown in the figure, let the center of the ellipse or circle corresponding to the arc trajectory formed by the wind of the sugar apple fruit be point P(0,y0), and let the critical points of the arc trajectory on the ellipse or circle be point A(x a ,y a ) and point B(x b ,y b ), let the coordinates of the location of the automated picking equipment be Q(x0,0), then the intersection of the straight line PQ and the partial circle is point R(R x ,R y ), let the moving speed v of the robotic arm of the automated picking equipment be L constant;

[0169] b. This application discusses the situation where the sugar apple fruit performs uniform and uniformly accelerated motion in a circle or ellipse. When the sugar apple fruit is blown by the wind on a circular arc trajectory to perform uniform circular motion and uniformly accelerated circular motion, the movement speed of the robotic arm is calculated based on the condition that the time it takes for the sugar apple fruit to move to point R is the same as the time it takes for the robotic arm to move to point R.

[0170] The following further explains the calculation process for obtaining the robot arm's moving speed when the sugar apple fruit is moving at a uniform speed or uniform acceleration in a circle or ellipse. Specifically:

[0171] b.1. When the sugar apple fruit is blown by the wind in uniform circular motion, the robot arm's movement speed is:

[0172]

[0173] Where L is the distance from the robot arm to the center of the circle or ellipse, and t1 is the movement time. However, since t1 is unknown, the movement speed of the robot arm can be obtained by assuming that the time it takes for the sugar apple fruit to move to point R is the same as the time it takes for the robot arm to move to point R. Since the sugar apple fruit performs uniform circular motion, the angular velocity of the sugar apple fruit is ω. ab Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0174]

[0175] Among them, θ a ,θ b are the angles of the sugar apple fruit when it moves to point A and point B respectively. According to the knowledge of polar coordinates,

[0176] Construct the parametric expression for the motion of the sugar apple fruit on the circular trajectory:

[0177]

[0178] Where x(t) and y(t) are the horizontal and vertical coordinates of the position of the sugar apple fruit at the time t, r is the radius of the circle on which the sugar apple fruit moves, and ω = ω ab , y0 is the vertical coordinate of the center of the circle where the sugar apple fruit moves on the arc trajectory, is a constant.

[0179] Point R(R x ,R y ) is the intersection point of the straight line PQ on the partial circle, so:

[0180]

[0181] After calculation, we can get:

[0182]

[0183] Set point R(R x ,R y ) into the parameter expression, and the time it takes for the sugar apple fruit to move from point A to point R is Then we have:

[0184]

[0185] After calculation, we can get:

[0186]

[0187] Since the sugar apple fruit can be divided into two situations during the movement process, the movement process of the sugar apple fruit on the arc trajectory is judged. The process of the sugar apple fruit moving from point A to point B or from point B to point A on the arc trajectory is called the sugar apple fruit moving on the arc trajectory once; if the sugar apple fruit and the robotic arm arrive at point R at the same time when the sugar apple fruit moves on the arc trajectory for the nth time, then when n = 2k-1, since the sugar apple fruit just moves from point A to point R, the time of the sugar apple fruit movement at this time is When n = 2k, since the sugar apple fruit just moves from point B to point R, the time it takes for the sugar apple fruit to move is Among them, n and k are positive numbers, and k is not equal to 0; so the movement time of the sugar apple fruit is expressed as:

[0188]

[0189] Based on the movement time t1 of the sugar apple fruit, the moving speed v of the robot arm is obtained. L :

[0190]

[0191] or

[0192]

[0193] b.2. When the sugar apple fruit is blown by the wind in uniformly accelerated circular motion, assuming that the angular acceleration α of the sugar apple fruit is constant:

[0194]

[0195] Among them, a t is the tangential acceleration of the sugar apple when it performs uniformly accelerated circular motion on the circle, and r is the radius of the circle on which the sugar apple moves on the cylindrical trajectory;

[0196] According to the velocity formula of uniformly accelerated circular motion v(t) = v0 + a t t, the speed of the sugar apple fruit moving from point A to point B is:

[0197]

[0198] According to the movement speed v of the sugar apple fruit at point A and point B A ,v B Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0199]

[0200] Among them, v A ,v B ≥0.

[0201] Construct the parametric expression for the motion of the sugar apple fruit on the circular trajectory:

[0202]

[0203] in θ0 and ω0 are constants. θ0 can be calculated because when the sugar apple fruit starts moving from point A, substituting time t = 0 into the expression for x(t) yields:

[0204]

[0205] Similarly, point R(R x ,R y ) into the parameter expression, and the time it takes for the sugar apple fruit to move from point A to point R is Then we have:

[0206]

[0207] After calculation, we can get:

[0208]

[0209] Negative terms have been discarded, so naturally Obviously, it is true; please note the calculation process here:

[0210]

[0211] Just like uniform circular motion, if the sugar apple fruit moves on the arc trajectory for the nth time, the sugar apple fruit and the robotic arm arrive at point R at the same time. When n = 2k-1, since the sugar apple fruit just moves from point A to point R, the movement time of the sugar apple fruit is When n = 2k, since the sugar apple fruit just moves from point B to point R, the movement time of the sugar apple fruit is So the motion time of the sugar apple fruit is expressed as:

[0212]

[0213] Based on the movement time t2 of the sugar apple fruit, the moving speed v of the robot arm is obtained. L :

[0214]

[0215] or

[0216]

[0217] b.3. When the sugar apple fruit is blown by the wind in uniform circular motion on an ellipse, let the angular velocity of the sugar apple be ω. ab Calculate the time it takes for the sugar apple fruit to move from point A to point B

[0218]

[0219] Among them, θ a ,θ b are the angles of the sugar apple fruit when it moves to point A and point B respectively;

[0220] Construct the parametric expression for the motion of the sugar apple fruit on the circular trajectory:

[0221]

[0222] Where x(t) and y(t) are the horizontal and vertical coordinates of the custard apple fruit when it moves on the arc trajectory for time t, a is the major axis of the ellipse, b is the minor axis of the ellipse, and ω = ω ab , y0 is the vertical coordinate of the center of the ellipse, is a constant.

[0223] Set point R(R x ,R y ) into the parameter expression, and the time it takes for the sugar apple fruit to move from point A to point R is Then we have:

[0224]

[0225] After calculation, we can get:

[0226]

[0227] Similarly, if the sugar apple and the robot arm arrive at point R at the same time when the sugar apple moves in the arc trajectory for the nth time, when n = 2k-1, since the sugar apple just moves from point A to point R, the movement time of the sugar apple is When n = 2k, since the sugar apple fruit just moves from point B to point R, the movement time of the sugar apple fruit is So the movement time of the sugar apple fruit is:

[0228]

[0229] Based on the movement time t3 of the sugar apple fruit, the moving speed v of the robot arm is obtained. L :

[0230]

[0231] or

[0232]

[0233] b.4. When the sugar apple fruit is blown by the wind in uniformly accelerated circular motion on an ellipse, obtain the tangential velocity a of the sugar apple fruit. t :

[0234]

[0235] Where a is the length of the major axis of the ellipse, b is the length of the minor axis of the ellipse, and θ is the angle of the sugar apple fruit when it moves on the circular trajectory;

[0236] Construct a parametric expression for the motion of the sugar apple fruit on a partial circle:

[0237]

[0238] Set point R(R x ,R y ) into the parameter expression, and the time it takes for the sugar apple fruit to move from point A to point R is Then we have:

[0239]

[0240] After calculation, we can get:

[0241]

[0242] Similarly, to judge the motion process of the sugar apple fruit on the circular trajectory, if the sugar apple fruit and the robotic arm arrive at point R at the same time during the nth partial circle motion, when n = 2k-1, since the sugar apple fruit just moves from point A to point R, the motion time of the sugar apple fruit at this time is When n = 2k, since the sugar apple fruit just moves from point B to point R, the movement time of the sugar apple fruit is So the movement time of the sugar apple fruit is:

[0243]

[0244] The moving speed v of the robotic arm is calculated based on the movement time t4 of the sugar apple fruit L .

[0245]

[0246] or

[0247]

[0248] For the communication module, it uses low-latency 5G communication technology to be responsible for data transmission and information sharing between modules. This module uses low-latency 5G communication technology to achieve ultra-high-speed, low-latency data interaction between individual vision modules, orchard vision modules, control modules, auxiliary modules, etc., to ensure synchronous data updates and rapid response to instructions. The high bandwidth and low latency characteristics of the 5G network enable automated picking equipment to receive real-time information such as fruit location, maturity, pest and disease detection, and quickly adjust picking strategies. In addition, the module can efficiently process data obtained from various sensors and computing units, and transmit it to the central computing module for real-time analysis. At the same time, the adjusted control instructions are sent to the execution module in milliseconds, ensuring high-speed coordination, precise control and dynamic adjustment of the entire system, improving the stability and efficiency of automated picking, and maintaining efficient operations even in large-scale orchard environments.

[0249] The control module is used to control and coordinate the work between modules to ensure the smooth execution of various tasks. Based on the decisions of the central processing module, it controls the movement of the robotic arm, the shooting of the vision module, and the spraying of medicine. The control module receives data from the individual vision module and the orchard vision module in real time, analyzes the maturity of the sugar apple fruit, the pest and disease situation, and the changes in the orchard environment, generates corresponding control instructions, and optimizes the picking path and medicine spraying strategy. Through precise instruction transmission, the control module ensures the efficient operation and precise operation of the automated picking equipment, thereby improving the efficiency and accuracy of sugar apple picking.

[0250] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0251] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A machine vision-based sugar apple intelligent cultivation and energy-saving collection system, characterized in that: The system includes an orchard vision module, an individual vision module, a control module, an auxiliary module, a central processing module and a communication module; The control module is connected to the orchard vision module, the individual vision module, the central processing module and the communication module respectively; the auxiliary module is connected to the orchard vision module and the individual vision module respectively; The orchard vision module uses ground-based fixed cameras and aerial drones to accurately locate and model the sugar apple orchard using computer vision technology and three-dimensional reconstruction algorithms, thereby obtaining the three-dimensional spatial structure of the sugar apple orchard and the location and distribution of the sugar apple fruits. The individual vision module is mounted on the automated picking equipment, and uses a first target detection algorithm to detect the maturity of the sugar apple fruit, and uses a second target detection algorithm to identify pests and diseases on the sugar apple and sugar apple leaves; The auxiliary module includes a visual enhancement module and an exposure optimization processor; The visual enhancement module enhances the collected visual image based on the illumination sensor data of the orchard visual module or the individual visual module; The exposure optimization processor is used to perform intelligent exposure control on the illumination of the orchard vision module or the individual vision module; The central processing module calculates the speed at which the robotic arm of the automated picking device just picks the sugar-apple fruit based on the motion trajectory of the sugar-apple fruit blown by the wind; the motion trajectory of the sugar-apple fruit blown by the wind is an ellipse or an arc on a circle; The communication module uses low-latency 5G communication technology to be responsible for data transmission and information sharing between modules; The control module is used to control and coordinate the work among the modules.

2. The sugar apple intelligent cultivation and energy-saving collection system according to claim 1, characterized in that: The ground fixed camera device is a high-resolution camera or a laser radar, which is used to collect ground data of the sugar apple orchard; the ground data is image and depth data of a local area of ​​the sugar apple orchard; The aerial drone captures aerial data of the sugar apple orchard from the air and generates three-dimensional point cloud data through structured light technology or stereo vision; the aerial data is a high-definition image of the sugar apple orchard; The orchard vision module integrates ground data and aerial data, and applies SLAM technology and deep learning algorithms to reconstruct the three-dimensional spatial structure of the sugar apple orchard and identify the location and distribution of sugar apple fruits.

3. The sugar apple intelligent cultivation and energy-saving collection system according to claim 1, characterized in that: The visual enhancement module enhances the collected visual image based on the light sensing data of the orchard visual module or the individual visual module, specifically: An environmental quality evaluation model is constructed based on light sensor data and image feature analysis, and the environmental state function is defined: Where Φ is the environmental quality evaluation model, L is the representation value of light intensity after nonlinear mapping, C is the fog concentration index based on dark channel statistics, N is the noise assessment value combined with frequency domain analysis and light sensor readings, γ is the noise sensitivity coefficient corresponding to the noise assessment value N in the environmental quality score, α(t) and β(t) are dynamic weight coefficients, which are adaptively adjusted according to the circadian rhythm to meet the following requirements: k1 is the light weight adjustment rate coefficient, L target The target light level set for the model, L current The real-time light level measurement value of the light sensor; Construct a decision function and use the dark light enhancement branch to perform dark light enhancement on the collected visual image or optimize the transmittance for haze interference based on the overall image brightness and the real-time light level measurement value of the light sensor to obtain a basic enhanced image; A gradient domain blending strategy is used to fuse high-frequency details with basic enhancement results to obtain a fused enhanced image. A reflection suppression model is constructed to suppress reflections on the fused enhanced image to obtain a visually enhanced image.

4. The sugar apple intelligent cultivation and energy-saving collection system according to claim 3, characterized in that: When the overall brightness of the visual image is lower than the set brightness threshold, and the real-time light level measurement value of the light sensor is lower than the set light level threshold, the environment of the visual image is judged to be a dark light scene. When the environmental state function Q is less than the lower threshold θ of the environmental state function Q, low When , the decision function uses the dark light enhancement branch to perform dark light enhancement on the collected visual image, specifically: A dual-channel fusion model is introduced to enhance dark light of the collected visual images: E(x,y)=ω·T cnn (I)+(1-ω)·R msr (I), Among them, (x, y) is the coordinate of a pixel in the captured image, T cnn is the enhanced result of the lightweight convolutional network output, R msr To improve the output of the multi-scale Retinex algorithm; ω is the fusion weight, which is dynamically calculated by the local contrast σ(x,y) of the image: k2 is the sensitivity coefficient that controls the speed at which the fusion weight ω changes with the local contrast, and σ0 is the set local contrast threshold.

5. The sugar apple intelligent cultivation and energy-saving collection system according to claim 3, characterized in that: When the overall brightness of the visual image is within the normal brightness range but there are obvious low-frequency blur features, and the dark channel map statistics show that the fog concentration is higher than the set fog concentration threshold, the visual image environment is judged to be a fog and haze interference environment. The decision function optimizes the transmittance for fog and haze interference, specifically: Define the corrected transmittance estimation function: t′(x)=max(t(x),η·G(x)⊙M fruit ), Among them, G(x) is the regional growth function of the sugar apple fruit, M fruit is the probability map of the sugar apple fruit image output by the mask generator, η is the canopy penetration compensation factor, and ⊙ is the Hadamard product; Use the corrected transmittance estimation function to restore the visual image and obtain a restored clear image; The texture preservation constraint term is introduced to optimize the modified transmittance estimation function: Where I is the visual image, J is the estimated value of the restored clear image, t′ is the corrected transmittance estimation function, A() is the estimated value of the global background light, λ is the weight coefficient of the texture preservation constraint term, S texture is the image texture guide map.

6. The sugar apple intelligent cultivation and energy-saving collection system according to claim 3, characterized in that: The high-frequency details are extracted by applying high-pass filtering or edge detection to the visual image; the fusion process is expressed as: Among them, J final is the fusion enhanced image, μ is the adjustment coefficient of the fusion enhancement strength, sign(D) is used to retain the change direction of the high-frequency details D. If the sign() input is a positive number, it returns +1; if the sign() input is a negative number, it returns -1; if the sign() input is zero, it returns 0; The reflection suppression model is expressed as: Where R suppress is the reflection suppression model, is the brightness channel of the fused enhanced image in the HSV color space, ρ is the peel reflection threshold of the sugar apple fruit, and n is the nonlinear suppression strength coefficient.

7. The sugar apple intelligent cultivation and energy-saving collection system according to claim 1, characterized in that: The exposure optimization processor is used to perform intelligent exposure control on the illumination of the orchard vision module or the individual vision module, specifically: Based on the spectral characteristics of photosynthetically active radiation, an exposure quality evaluation model is constructed, and the pixel-level exposure anomaly index is defined: Where Q exp (x) is the pixel-level exposure anomaly index at pixel x in the visual image collected by the orchard vision module or the individual vision module, L(x) is the light intensity value of pixel x, V(x) is the chlorophyll fluorescence intensity value at pixel x, σ v is the standard deviation of chlorophyll fluorescence intensity, V opt is the optimal value of chlorophyll fluorescence intensity, μ PAR is the expected value of photosynthetically active radiation, which is dynamically updated by the chlorophyll fluorescence dynamics model: in, is the biomass change rate of the plant, F(t) is the chlorophyll fluorescence intensity at time t, and F max is the maximum fluorescence intensity value under light saturation state, k p is the rate coefficient of photosynthetic active radiation renewal; Design a dual-channel nonlinear transfer function to process overexposed and underexposed areas separately: Among them, T(I) is the enhanced result of the output visual image, I γ(t) is the image after dynamic gamma correction, γ(t) is the dynamic gamma correction parameter, β is the sensitivity coefficient of brightness adjustment, I is the brightness value of the input visual image, I th is the threshold value of overexposure and underexposure areas, and the dynamic adjustment parameters γ(t) and α(t) are constrained by photosynthetic efficiency: is the plant biomass change rate, is the sensitivity of chlorophyll fluorescence intensity to light intensity, k f is the sensitivity coefficient of photosynthesis to light intensity, F0 is the initial fluorescence intensity of photosynthesis; In response to the need for continuous data collection in agricultural monitoring, a time series smoothing constraint based on metabolic rhythm is proposed: Among them, I t is the image at time t, I t-1 is the image at time t-1, ||·|| W is the weighted Euclidean distance, is the gradient of the pixel-level exposure anomaly index at time t; By fusing multispectral sensor data with RGB image features, the light intensity distribution in the photosynthetically active radiation band is reconstructed: Among them, I PAR is the light intensity distribution in the photosynthetically active radiation band, λ is the photosynthetically active radiation band, PAR λ is the radiation intensity in the photosynthetically active radiation band, I RGB is the image collected by the orchard vision module or the individual vision module, ξ is the scaling factor, and the weight function w(λ) is determined by the gradient derived from the chlorophyll absorption spectrum φ(λ): is the derivative of the weight function with respect to the photosynthetically active radiation band, k w is the weight coefficient, is the gradient of the chlorophyll absorption spectrum, is the rate of change of plant growth rate over time; CNN spec It is a lightweight spectral prediction network used to estimate PAR components from RGB images when there is no multispectral sensor.

8. The sugar apple intelligent cultivation and energy-saving collection system according to claim 1, characterized in that: The speed at which the robotic arm of the automated picking device just picks the sugar-apple fruit is calculated based on the motion trajectory of the sugar-apple fruit blown by the wind, specifically: Let the center of the ellipse or circle corresponding to the arc trajectory formed by the wind of the sugar apple fruit be point P, let the critical points of the arc trajectory on the ellipse or circle be points A and B respectively, let the coordinates of the position of the automated picking equipment be Q, then the intersection of the straight line PQ and the partial circle is point R, let the moving speed of the robotic arm of the automated picking equipment be v L constant; When the sugar apple fruit is blown by the wind on an ellipse or circle to perform uniform circular motion and uniform accelerated circular motion, the moving speed of the robotic arm is calculated based on the condition that the time it takes for the sugar apple fruit to move to point R is the same as the time it takes for the robotic arm to move to point R.

9. The sugar apple intelligent cultivation and energy-saving collection system according to claim 8, characterized in that: The calculation obtains the moving speed of the robotic arm, specifically: When the sugar apple fruit is blown by the wind and performs uniform circular motion on the circle, let the angular velocity of the sugar apple fruit be ω ab Calculate the time it takes for the sugar apple fruit to move from point A to point B Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R. Determine the movement process of the sugar apple fruit on the circular trajectory. Assume that the process of the sugar apple fruit moving from point A to point B or from point B to point A on the circular trajectory is one movement of the sugar apple fruit on the circular trajectory. If the sugar apple fruit and the robotic arm arrive at point R at the same time when the sugar apple fruit moves on the circular trajectory for the nth time, when n = 2k-1, the sugar apple fruit just moves from point A to point R. Calculate the movement time of the sugar apple fruit. When n=2k, the sugar apple fruit just moves from point B to point R. Calculate the movement time of the sugar apple fruit. Where n and k are positive numbers, and k is not equal to 0; The motion time of the sugar apple fruit is expressed as: The moving speed v of the robotic arm is calculated based on the movement time t1 of the sugar apple fruit L ; When the sugar apple fruit is blown by the wind on a circle to perform uniform accelerated circular motion, assuming that the angular acceleration α of the sugar apple fruit is constant, according to the speed v of the sugar apple fruit at point A and point B, A ,v B Calculate the time it takes for the sugar apple fruit to move from point A to point B Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R. Similarly, the motion process of the sugar apple fruit on the arc trajectory is determined to obtain the motion time t2 of the sugar apple fruit as: The moving speed v of the robotic arm is calculated based on the movement time t2 of the sugar apple fruit L ; When the sugar apple fruit is blown by the wind and performs uniform circular motion on the ellipse, let the angular velocity of the sugar apple fruit be ω ab Calculate the time it takes for the sugar apple fruit to move from point A to point B Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move to point R. Determine the motion process of the sugar apple fruit on the arc trajectory. Similarly, the motion time t3 of the sugar apple fruit is obtained as: The moving speed v of the robotic arm is calculated based on the movement time t3 of the sugar apple fruit L ; When the sugar apple fruit is blown by the wind and performs uniformly accelerated circular motion on the ellipse, the tangential velocity a of the sugar apple fruit is obtained. t Calculate the time it takes for the sugar apple fruit to move from point A to point B Construct a parametric expression for the motion of the sugar apple fruit on the circular trajectory, and substitute point R into the parametric expression to obtain the time it takes for the sugar apple fruit to move from point A to point R. Similarly, the movement process of the sugar apple fruit on the arc trajectory is judged, and the movement time t4 of the sugar apple fruit is obtained as: The moving speed v of the robotic arm is calculated based on the movement time t4 of the sugar apple fruit L .

10. The sugar apple intelligent cultivation and energy-saving collection system according to claim 1, characterized in that: The individual vision module is composed of a camera and an intelligent chip; The pests and diseases include anthracnose, blank canker, beam rot, fruit leaf spot, leaf spot and whitefly; The automated picking equipment is equipped with a medicine spraying facility; when the second target detection algorithm of the individual visual module identifies pests and diseases on the sugar apple and sugar apple leaves, the corresponding medicine is sprayed to treat and eliminate the pests.

Images