A method for controlling an indoor landscape environment

Abstract

This invention discloses an indoor landscape environment control method, belonging to the field of image processing technology. This invention acquires indoor landscape images, extracts G channel feature values ​​based on the RGB channel value distribution, constructs a green distribution map, and divides it into blocks. On the one hand, this reduces the amount of data processed by the model, and on the other hand, it facilitates regional classification, finds vegetation areas, and calculates the density of landscape vegetation. Combined with the area of ​​the vegetation area, the landscape lighting is precisely controlled, realizing a method for precisely controlling landscape light intensity based on the density of landscape vegetation and the area of ​​the vegetation area.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to an indoor landscape environment control method. Background Technology

[0002] Indoor landscaping refers to bringing outdoor natural and man-made scenery directly indoors to create a harmonious and pleasant space. In most cases, vegetation is chosen as the landscaping material. However, indoor plants cannot receive sufficient sunlight, so artificial lighting is designed to compensate for the lack of natural light. This typically involves using specific lighting equipment, such as fluorescent lamps or LED lights, to simulate the effect of natural light and ensure healthy plant growth. However, different plant growth conditions require different amounts of light. Current technology only uses lighting equipment to compensate for the lack of light, but both excessively weak and excessive light can damage plants. In cases of excessive light intensity, the plant's photosynthetic system is damaged, reducing photosynthetic efficiency. Furthermore, when light exceeds the photosynthetic saturation point, excess light energy may be converted into heat, leading to increased temperature and negatively impacting the plant. In cases of insufficient light intensity, chlorophyll synthesis is hindered, resulting in poor plant growth. Appropriate light intensity is crucial for the healthy growth of plants. Both excessively weak and excessively strong light intensity can affect the normal growth and development of vegetation through various pathways. Therefore, it is necessary to adjust the light conditions according to the specific conditions of the vegetation to ensure that it can grow healthily. Summary of the Invention

[0003] In view of the above-mentioned shortcomings in the prior art, the present invention provides an indoor landscape environment control method that solves the problem of inaccurate light intensity control in indoor landscapes.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is: an indoor landscape environment control method, comprising the following steps:

[0005] S1. Acquire indoor landscape images;

[0006] S2. Extract the green distribution map from the indoor landscape image based on the RGB channel value distribution;

[0007] S3. Divide the green distribution map into blocks to obtain multiple image blocks;

[0008] S4. Each image block is processed using a vegetation region recognition model to obtain a vegetation region, wherein the vegetation region consists of multiple image blocks that are identified as vegetation.

[0009] S5. Calculate the density of landscape vegetation based on the vegetation area;

[0010] S6. Control the landscape lighting according to the density of the landscape vegetation and the area of ​​the vegetation zone.

[0011] The beneficial effects of this invention are as follows: This invention acquires indoor landscape images, extracts G channel feature values ​​based on the RGB channel value distribution, constructs a green distribution map, and divides it into blocks. On the one hand, this reduces the amount of data processed by the model, and on the other hand, it facilitates regional classification, identifies vegetation areas, calculates the density of landscape vegetation, and combines the area of ​​vegetation areas to precisely control landscape lighting, thus realizing a method for precisely controlling landscape light intensity based on the density of landscape vegetation and the area of ​​vegetation areas.

[0012] Furthermore, step S2 includes the following sub-steps:

[0013] S21. Obtain the green factor of each pixel based on the RGB channel values ​​of the indoor landscape image;

[0014] S22. Based on the RGB channel values ​​of the indoor landscape image, obtain the G channel scaling factor for each pixel.

[0015] S23. Adjust the G channel scaling factor of each pixel using the green factor of each pixel to obtain a green distribution map.

[0016] Furthermore, the expression for the green factor of each pixel in S21 is:

[0017] ,

[0018] Where, γ i Let R be the green factor of the i-th pixel. i G is the R channel value of the i-th pixel in the indoor landscape image. i Let B be the G channel value of the i-th pixel in the indoor landscape image. i Let be the B channel value of the i-th pixel in the indoor landscape image, where i is a positive integer and || is the absolute value operation.

[0019] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, the distance between the G channel value at each pixel location and the cube of the R channel value and the B channel value is measured. When the green factor of the pixel is positive, the G channel value is significant, and when the green factor of the pixel is negative, the R channel value and the B channel value have a high proportion.

[0020] Furthermore, the expression for the G channel scaling factor of each pixel in S22 is as follows:

[0021] ,

[0022] Among them, G p,i is the scaling factor of the G channel at the i-th pixel.

[0023] The beneficial effects of the above-mentioned further solution are as follows: The present invention extracts the G channel ratio coefficient to characterize the proportion of the G channel value at each pixel. The larger the proportion, the more prominent the green, the more chloroplasts in the plant, and the more prominent the green ratio.

[0024] Furthermore, the expression for adjustment in S23 is:

[0025] ,

[0026] Among them, G d,i Let be the green distribution value of the i-th pixel on the green distribution map.

[0027] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, when the green factor of a pixel is positive, the G channel value is higher than the R channel value and the B channel value, which enhances the G channel scaling factor and further highlights the green feature. When the green factor of a pixel is negative, the R channel value and the B channel value are higher than the G channel value, which weakens the G channel scaling factor, making the green feature more prominent and the non-green feature weaker, thereby improving the model's classification ability.

[0028] Furthermore, the vegetation region identification model in S4 includes: a first convolutional layer, a second convolutional layer, a ReLU layer, a first feature extraction channel, a second feature extraction channel, a third feature extraction channel, a Concat layer, an Avgpool layer, a Maxpool layer, a first classification layer, a second classification layer, and an output layer;

[0029] The input of the first convolutional layer serves as the input of the vegetation region identification model, and its output is connected to the input of the ReLU layer. The output of the ReLU layer is connected to the inputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively. The input of the Concat layer is connected to the outputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively, and its output is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the inputs of the Avgpool layer and the Maxpool layer, respectively. The input of the first classification layer is connected to the output of the Avgpool layer. The input of the second classification layer is connected to the output of the Maxpool layer. The input of the output layer is connected to the outputs of the first and second classification layers, respectively, and its output serves as the output of the vegetation region identification model.

[0030] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, three feature extraction channels are set up to extract different features respectively. Feature fusion is performed at the Concat layer. The Maxpool layer is used to extract the maximum feature value of the window. The first classification layer is used to classify based on each maximum feature value. The Avgpool layer is used to extract the average feature value of the window. The second classification layer is used to classify based on each average feature value. The output layer integrates the two classification results to obtain the final classification, thereby improving the classification accuracy.

[0031] Furthermore, the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel each include: a third convolutional layer, a fourth convolutional layer, a first Tanh layer, a second Tanh layer, a first multiplier M1, and a second multiplier M2;

[0032] The input of the third convolutional layer serves as the input of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. Its output is connected to the input of the first Tanh layer and the first input of the first multiplier M1, respectively. The second input of the first multiplier M1 is connected to the output of the first Tanh layer, and its output is connected to the input of the fourth convolutional layer. The output of the fourth convolutional layer is connected to the input of the second Tanh layer and the first input of the second multiplier M2, respectively. The second input of the second multiplier M2 is connected to the output of the second Tanh layer, and its output serves as the output of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. The kernel size of the third convolutional layer in the first feature extraction channel is... The kernel size of the third convolutional layer in the second feature extraction channel is The kernel size of the third convolutional layer in the third feature extraction channel is .

[0033] The beneficial effects of the above-mentioned further scheme are as follows: The feature extraction channel of the present invention is provided with two Tanh layers and two multipliers. Each Tanh layer is used to calculate the weight of the feature value input to the Tanh layer, and each multiplier is used to multiply the input and output of the Tanh layer to realize feature self-enhancement. The use of double-layer self-enhancement processing can significantly improve the attention of salient features and improve classification accuracy.

[0034] Furthermore, the formula for calculating the density of landscape vegetation in S5 is as follows:

[0035] ,

[0036] Where y represents the density of landscape vegetation, and G d,j,k Let M be the green distribution value of the k-th pixel in the j-th image patch of the vegetation region, and M be the number of image patches in the vegetation region. jN represents the number of pixels in the j-th image patch. j,d Let be the number of pixels in the j-th image patch whose green distribution value is greater than the threshold, where j and k are positive integers.

[0037] The beneficial effects of the above-mentioned further solution are as follows: In this invention, the density of landscape vegetation is considered from two aspects. One aspect is the green distribution value on each image block, and the other aspect is the number of pixels on the image block whose green distribution value is greater than the threshold. The more pixels whose green distribution value is greater than the threshold, the higher the density of such pixels, and thus the greater the density of landscape vegetation.

[0038] Furthermore, step S6 includes the following sub-steps:

[0039] S61. Calculate the target light intensity based on the density of the landscape vegetation and the area of ​​the vegetation region;

[0040] S62. Based on the target light intensity, a light control model is used to stably control the landscape lighting.

[0041] Furthermore, the formula for calculating the target illumination intensity in S61 is as follows:

[0042] ,

[0043] Among them, L tar Let y represent the target light intensity, y represent the density of landscape vegetation, and a represent the density of light intensity. rea Let w1 be the area of ​​the vegetated region, w2 be the first weight, and e be the natural constant.

[0044] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, the influence of both the density of landscape vegetation and the area of ​​vegetation area are considered simultaneously, and a mapping function from the density of landscape vegetation and the area of ​​vegetation area to the target light intensity is established, thereby achieving accurate estimation of the target light intensity.

[0045] Furthermore, the expression for the illumination control model in S62 is:

[0046] ,

[0047] Among them, I t Let I be the current of the landscape light at time t. t-1 Let I be the current of the landscape light at time t-1. s L is the current step size. tar For the target light intensity, L t-1 Let be the actual illumination intensity at time t-1, and arctan be the arctangent function.

[0048] The beneficial effects of the above-mentioned further solution are as follows: In this invention, the current of the landscape lamp is adjusted according to the distance between the actual light intensity and the target light intensity at the previous moment. When the target light intensity is greater than the actual light intensity, the current of the landscape lamp is increased, and when the actual light intensity is greater than the target light intensity, the current of the landscape lamp is decreased, so that the actual light intensity is always around the level of the target light intensity, so that the vegetation obtains an adaptive light intensity. Attached Figure Description

[0049] Figure 1 A flowchart of an indoor landscape environment control method;

[0050] Figure 2 A schematic diagram of the structure of a vegetation area identification model;

[0051] Figure 3 This is a schematic diagram of the structure of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel. Detailed Implementation

[0052] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0053] like Figure 1 As shown, an indoor landscape environment control method includes the following steps:

[0054] S1. Acquire indoor landscape images;

[0055] S2. Extract the green distribution map from the indoor landscape image based on the RGB channel value distribution;

[0056] S3. Divide the green distribution map into blocks to obtain multiple image blocks;

[0057] S4. Each image block is processed using a vegetation region recognition model to obtain a vegetation region, wherein the vegetation region consists of multiple image blocks that are identified as vegetation.

[0058] S5. Calculate the density of landscape vegetation based on the vegetation area;

[0059] S6. Control the landscape lighting according to the density of the landscape vegetation and the area of ​​the vegetation zone.

[0060] In this embodiment, when the green distribution map is segmented in step S3, the green distribution map is evenly divided into multiple image blocks, and the multiple image blocks can be stitched together to form the green distribution map.

[0061] S2 includes the following steps:

[0062] S21. Obtain the green factor of each pixel based on the RGB channel values ​​of the indoor landscape image;

[0063] S22. Based on the RGB channel values ​​of the indoor landscape image, obtain the G channel scaling factor for each pixel.

[0064] S23. Adjust the G channel scaling factor of each pixel using the green factor of each pixel to obtain a green distribution map.

[0065] The expression for the green factor of each pixel in S21 is:

[0066] ,

[0067] Where, γ i Let R be the green factor of the i-th pixel. i G is the R channel value of the i-th pixel in the indoor landscape image. i Let B be the G channel value of the i-th pixel in the indoor landscape image. i Let be the B channel value of the i-th pixel in the indoor landscape image, where i is a positive integer and || is the absolute value operation.

[0068] In this invention, the distance between the G channel value at each pixel location and the cube of the R channel value and the B channel value is measured. When the green factor of the pixel is positive, the G channel value is significant, and when the green factor of the pixel is negative, the R channel value and the B channel value account for a high proportion.

[0069] The expression for the G channel scaling factor of each pixel in S22 is:

[0070] ,

[0071] Among them, G p,i is the scaling factor of the G channel at the i-th pixel.

[0072] In this invention, the G channel ratio coefficient is extracted to characterize the proportion of the G channel value at each pixel. The larger the proportion, the more prominent the green, the more chloroplasts in the plant, and the more prominent the green ratio.

[0073] The expression for adjustment in S23 is:

[0074] ,

[0075] Among them, G d,i Let be the green distribution value of the i-th pixel on the green distribution map.

[0076] In this invention, when the green factor of a pixel is positive, the G channel value is higher than the R channel value and the B channel value, which enhances the G channel scaling factor and further highlights the green feature. When the green factor of a pixel is negative, the R channel value and the B channel value are higher than the G channel value, which weakens the G channel scaling factor. This results in a more prominent green feature and a weakening of non-green features, thereby improving the model's classification ability.

[0077] In this invention, the green distribution map is equivalent to replacing the original pixel value with a green distribution value at each pixel point of the indoor landscape image.

[0078] like Figure 2 As shown, the vegetation area identification model in S4 includes: a first convolutional layer, a second convolutional layer, a ReLU layer, a first feature extraction channel, a second feature extraction channel, a third feature extraction channel, a Concat layer, an Avgpool layer, a Maxpool layer, a first classification layer, a second classification layer, and an output layer;

[0079] The input of the first convolutional layer serves as the input of the vegetation region identification model, and its output is connected to the input of the ReLU layer. The output of the ReLU layer is connected to the inputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively. The input of the Concat layer is connected to the outputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively, and its output is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the inputs of the Avgpool layer and the Maxpool layer, respectively. The input of the first classification layer is connected to the output of the Avgpool layer. The input of the second classification layer is connected to the output of the Maxpool layer. The input of the output layer is connected to the outputs of the first and second classification layers, respectively, and its output serves as the output of the vegetation region identification model.

[0080] This invention sets up three feature extraction channels to extract different features respectively. Feature fusion is performed at the Concat layer. The Maxpool layer is used to extract the maximum feature value of the window. The first classification layer classifies based on each maximum feature value. The Avgpool layer is used to extract the average feature value of the window. The second classification layer classifies based on each average feature value. The output layer combines the two classification results to obtain the final classification, thereby improving the classification accuracy.

[0081] In this embodiment, the expressions for both the first classification layer and the second classification layer are: Where S is the output of the classification layer, x m ω is the m-th feature value input to the classification layer. m For the m-th eigenvalue x m The weight, b m For the m-th eigenvalue x m The bias is C, where C is the number of eigenvalues ​​and m is a positive integer.

[0082] The output layer uses the sigmoid activation function, specifically: Where h is the output of the output layer, sigmoid is the activation function, S1 is the output of the first classification layer, β1 is the weight of the output of the first classification layer, S2 is the output of the second classification layer, β2 is the weight of the output of the second classification layer, and when the output of the output layer is greater than or equal to 0.5, the image patch is vegetation, and when it is less than 0.5, the image patch is non-vegetation.

[0083] like Figure 3 As shown, the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel each include: a third convolutional layer, a fourth convolutional layer, a first Tanh layer, a second Tanh layer, a first multiplier M1, and a second multiplier M2;

[0084] The input of the third convolutional layer serves as the input of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. Its output is connected to the input of the first Tanh layer and the first input of the first multiplier M1, respectively. The second input of the first multiplier M1 is connected to the output of the first Tanh layer, and its output is connected to the input of the fourth convolutional layer. The output of the fourth convolutional layer is connected to the input of the second Tanh layer and the first input of the second multiplier M2, respectively. The second input of the second multiplier M2 is connected to the output of the second Tanh layer, and its output serves as the output of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. The kernel size of the third convolutional layer in the first feature extraction channel is... The kernel size of the third convolutional layer in the second feature extraction channel is The kernel size of the third convolutional layer in the third feature extraction channel is .

[0085] The feature extraction channel of this invention is configured with two Tanh layers and two multipliers. Each Tanh layer is used to calculate the weight of the feature values ​​input to the Tanh layer, and each multiplier is used to multiply the input and output of the Tanh layer to achieve feature self-enhancement. The use of double-layer self-enhancement processing can significantly improve the attention of salient features and improve classification accuracy.

[0086] The formula for calculating the density of landscape vegetation in S5 is as follows:

[0087] ,

[0088] Where y represents the density of landscape vegetation, and G d,j,k Let M be the green distribution value of the k-th pixel in the j-th image patch of the vegetation region, and M be the number of image patches in the vegetation region. j N represents the number of pixels in the j-th image patch. j,d Let be the number of pixels in the j-th image patch whose green distribution value is greater than the threshold, where j and k are positive integers.

[0089] In this invention, the density of landscape vegetation is considered from two aspects: one is the green distribution value on each image block, and the other is the number of pixels on the image block whose green distribution value is greater than a threshold. The more pixels whose green distribution value is greater than the threshold, the higher the density of such pixels, and thus the greater the density of landscape vegetation.

[0090] S6 includes the following sub-steps:

[0091] S61. Calculate the target light intensity based on the density of the landscape vegetation and the area of ​​the vegetation region;

[0092] S62. Based on the target light intensity, a light control model is used to stably control the landscape lighting.

[0093] The formula for calculating the target illumination intensity in S61 is as follows:

[0094] ,

[0095] Among them, L tar Let y represent the target light intensity, y represent the density of landscape vegetation, and a represent the density of light intensity. rea Let w1 be the area of ​​the vegetated region, w2 be the first weight, and e be the natural constant.

[0096] This invention considers the influence of both the density of landscape vegetation and the area of ​​the vegetation region, and establishes a mapping function from the density of landscape vegetation and the area of ​​the vegetation region to the target light intensity, thereby achieving accurate estimation of the target light intensity.

[0097] The expression for the illumination control model in S62 is:

[0098] ,

[0099] Among them, I t Let I be the current of the landscape light at time t. t-1 Let I be the current of the landscape light at time t-1.s L is the current step size. tar For the target light intensity, L t-1 Let be the actual illumination intensity at time t-1, and arctan be the arctangent function.

[0100] In this invention, the current of the landscape lights is adjusted based on the distance between the actual light intensity and the target light intensity at the previous moment. When the target light intensity is greater than the actual light intensity, the current of the landscape lights is increased, and when the actual light intensity is greater than the target light intensity, the current of the landscape lights is decreased, so that the actual light intensity is always around the level of the target light intensity, allowing the vegetation to obtain an adaptive light intensity.

[0101] This invention acquires indoor landscape images, extracts G channel feature values ​​based on the RGB channel value distribution, constructs a green distribution map, and divides it into blocks. This reduces the amount of data processed by the model and facilitates regional classification, identifies vegetation areas, and calculates the density of landscape vegetation. Combined with the area of ​​the vegetation area, the landscape lighting is precisely controlled, realizing a method for precisely controlling landscape light intensity based on the density of landscape vegetation and the area of ​​the vegetation area.

[0102] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling indoor landscape environment, characterized in that, Includes the following steps: S1. Acquire indoor landscape images; S2. Extract the green distribution map from the indoor landscape image based on the RGB channel value distribution; S3. Divide the green distribution map into blocks to obtain multiple image blocks; S4. Each image block is processed using a vegetation region recognition model to obtain a vegetation region, wherein the vegetation region consists of multiple image blocks that are identified as vegetation. S5. Calculate the density of landscape vegetation based on the vegetation area; S6. Control the landscape lighting according to the density of the landscape vegetation and the area of ​​the vegetation zone; S2 includes the following steps: S21. Obtain the green factor of each pixel based on the RGB channel values ​​of the indoor landscape image; S22. Based on the RGB channel values ​​of the indoor landscape image, obtain the G channel scaling factor for each pixel. S23. Adjust the G channel scaling factor of each pixel using the green factor of each pixel to obtain a green distribution map; The expression for the green factor of each pixel in S21 is: ， Where, γ i Let R be the green factor of the i-th pixel. i G is the R channel value of the i-th pixel in the indoor landscape image. i Let B be the G channel value of the i-th pixel in the indoor landscape image. i Let be the B channel value of the i-th pixel in the indoor landscape image, where i is a positive integer and || is the absolute value operation; The expression for the G channel scaling factor of each pixel in S22 is: ， Among them, G p,i The scaling factor for the G channel of the i-th pixel; The expression for adjustment in S23 is: ， Among them, G d,i Let be the green distribution value of the i-th pixel on the green distribution map; The formula for calculating the density of landscape vegetation in S5 is as follows: ， Where y represents the density of landscape vegetation, and G d,j,k Let M be the green distribution value of the k-th pixel in the j-th image patch of the vegetation region, and M be the number of image patches in the vegetation region. j N represents the number of pixels in the j-th image patch. j,d Let be the number of pixels in the j-th image patch whose green distribution value is greater than the threshold, where j and k are positive integers.

2. The indoor landscape environment control method according to claim 1, characterized in that, The vegetation region identification model in S4 includes: a first convolutional layer, a second convolutional layer, a ReLU layer, a first feature extraction channel, a second feature extraction channel, a third feature extraction channel, a Concat layer, an Avgpool layer, a Maxpool layer, a first classification layer, a second classification layer, and an output layer. The input of the first convolutional layer serves as the input of the vegetation region identification model, and its output is connected to the input of the ReLU layer. The output of the ReLU layer is connected to the inputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively. The input of the Concat layer is connected to the outputs of the first feature extraction channel, the second feature extraction channel, and the third feature extraction channel, respectively, and its output is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the inputs of the Avgpool layer and the Maxpool layer, respectively. The input of the first classification layer is connected to the output of the Avgpool layer. The input of the second classification layer is connected to the output of the Maxpool layer. The input of the output layer is connected to the outputs of the first and second classification layers, respectively, and its output serves as the output of the vegetation region identification model.

3. The indoor landscape environment control method according to claim 2, characterized in that, The first feature extraction channel, the second feature extraction channel, and the third feature extraction channel each include: a third convolutional layer, a fourth convolutional layer, a first Tanh layer, a second Tanh layer, a first multiplier M1, and a second multiplier M2; The input of the third convolutional layer serves as the input of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. Its output is connected to the input of the first Tanh layer and the first input of the first multiplier M1, respectively. The second input of the first multiplier M1 is connected to the output of the first Tanh layer, and its output is connected to the input of the fourth convolutional layer. The output of the fourth convolutional layer is connected to the input of the second Tanh layer and the first input of the second multiplier M2, respectively. The second input of the second multiplier M2 is connected to the output of the second Tanh layer, and its output serves as the output of the first feature extraction channel, the second feature extraction channel, or the third feature extraction channel. The kernel size of the third convolutional layer in the first feature extraction channel is... The kernel size of the third convolutional layer in the second feature extraction channel is The kernel size of the third convolutional layer in the third feature extraction channel is .

4. The indoor landscape environment control method according to claim 1, characterized in that, S6 includes the following sub-steps: S61. Calculate the target light intensity based on the density of the landscape vegetation and the area of ​​the vegetation region; S62. Based on the target light intensity, a light control model is used to stably control the landscape lighting.

5. The indoor landscape environment control method according to claim 4, characterized in that, The formula for calculating the target illumination intensity in S61 is as follows: ， Among them, L tar Let y represent the target light intensity, y represent the density of landscape vegetation, and a represent the density of light intensity. rea Let w1 be the area of ​​the vegetated region, w2 be the first weight, and e be the natural constant.

6. The indoor landscape environment control method according to claim 4, characterized in that, The expression for the illumination control model in S62 is: ， Among them, I t Let I be the current of the landscape light at time t. t-1 Let I be the current of the landscape light at time t-1. s L is the current step size. tar For the target light intensity, L t-1 Let be the actual illumination intensity at time t-1, and arctan be the arctangent function.

Images