Agricultural crop state identification method and system based on image analysis
Through multi-wheel flight and multi-angle acquisition of drones, color information is used to identify crop states, which solves the problem of insufficient generalization ability of identification models in the prior art, and achieves more accurate and efficient crop identification.
Patent Information
- Application Number
- CN202410945731.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the identification model of agricultural and plant crops cannot fully cover various factors in the region, climate, crop species and growth stage, resulting in inaccurate identification results and insufficient generalization ability.
Using image analysis method, the target image is obtained through drones, color blocks are identified, and multi-wheel flights and multi-angle acquisition is carried out. The state of the crops is recognized by using color information to reduce data processing, improve identification efficiency, and is applicable. in different regions and crop types.
It improves the accuracy and efficiency of plant crop identification, reduces the impact on region, climate and crop types, and can obtain more comprehensive status information of plant crops, which is suitable for different identification situations.
Smart Images

Figure CN119007036A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image processing of agricultural crops, and in particular relates to an agricultural crop state recognition method and system based on image analysis. Background Art
[0002] With the development of image recognition technology, drones or satellite remote sensing can be used to collect images of crop planting areas, and then the status of crops in the planting areas can be identified based on the collected images to assist users in planting crops efficiently. However, in the existing technology, crops are mainly identified by pre-trained neural network models. However, the above methods need to take into account that image data in the agricultural field may be affected by multiple factors such as region, climate, crop type, and growth stage, resulting in the existing data sets being difficult to fully cover all possible situations and the recognition results being inaccurate due to insufficient model generalization capabilities. Summary of the invention
[0003] In view of the above problems, the purpose of the present invention is to disclose a method for identifying the status of agricultural crops based on image analysis. The method obtains the status of agricultural crops by identifying colors. The target image can be obtained by high-altitude photography, and the target image can be divided into regions to obtain abnormal areas. Then, at least one flight is performed on the abnormal areas to obtain corresponding judgment results. As mentioned above, the present invention uses multiple rounds of image acquisition and identifies the crop status based on color. It can be applicable to the identification of agricultural crops in different regions, climates, crop types, and growth stages, and can be applicable to different specific identification situations while ensuring the recognition accuracy.
[0004] The present invention includes:
[0005] A method for identifying the state of agricultural crops based on image analysis comprises the following steps: S100, obtaining a target image taken by a drone of a target area, wherein at least a part of the target area is planted with agricultural crops, and the target image is a color image; S200, obtaining pixel information of the target image as first pixel information, inputting the first pixel information into a classification recognition model to obtain a classification result, wherein the classification result includes a normal result and an abnormal result; S300, generating a mask according to the classification result, operating and processing the mask to divide each color block in the target image, wherein the color block is divided into at least a normal block and an abnormal block; S400, locating the position of the abnormal block in the target area as a sub-target area; S500, obtaining a sub-target image taken by a drone of the sub-target area; S600, obtaining pixel information containing color of the sub-target image as second pixel information, and comparing the second pixel information with a preset color range to obtain a color determination result.
[0006] As some preferred schemes, in S100, the drone performs a first flight to acquire the target image, and when performing the first flight, the coordinate information of the target area is recorded and the coordinate information is projected into the target image; in S500, the drone performs a second flight to acquire the sub-target image, and when performing the second flight, the drone reads the coordinate information corresponding to the sub-target area for: acquiring the sub-target image within the range of the sub-target area; wherein, the altitude at which the drone performs the first flight is higher than the altitude of the second flight.
[0007] As some preferred schemes, it also includes S610 executed after S600; S610, the color judgment result includes at least normal color and abnormal color; the color judgment result is obtained, the proportion of abnormal color in the color judgment result is calculated as a first proportion value, and the first proportion value is compared with a first preset threshold value to output a first judgment result, and the first judgment result includes normal crop and abnormal crop.
[0008] As some preferred schemes, it also includes S710-S720 performed after S610; S710, the UAV performs a third flight; when performing the third flight, an image of any single agricultural crop target in the abnormal block is obtained as a first individual image; S720, pixel information containing color of the first individual image is obtained as third pixel information, and the third pixel information is compared with the preset color range to obtain color distribution.
[0009] As some preferred schemes, it also includes S730-S750 performed after S720; S730, obtaining the color distribution, the color distribution at least includes the first color and the second color distributed in the first individual image, calculating the proportion of the first color to the second color as a second ratio value, comparing the second ratio value with a second preset threshold value and outputting a second judgment result, the second judgment result includes a single crop being normal and a single crop being abnormal; S740, according to the second judgment result, recording the number of results of the single crop being normal, and calculating the confidence interval of the single crop being normal; S750, comparing the first ratio value with the confidence interval; when the first ratio value falls within the confidence interval, outputting the first judgment result and / or the second judgment result; when the first ratio value does not fall within the confidence interval, increasing the first preset threshold by a first increase value and re-performing S610.
[0010] As some preferred schemes, the agricultural crops are fruits and vegetables, the single agricultural crop target is the leaves of fruits and vegetables, and the first individual image is a front view of the leaves; the second judgment result includes: a first sub-result, including normal leaves and abnormal leaves; a second sub-result, including green leaves, yellow leaves, dark green leaves, and white leaves.
[0011] As some preferred schemes, it also includes S711 executed synchronously with S710; S711, when the UAV performs the third flight, obtains a second individual image of any single agricultural crop target in the abnormal block; the second individual image is an image of the side of the leaf.
[0012] As some preferred schemes, it also includes S712 executed after S711; S712, inputting the second individual image into the first recognition model to obtain the radius estimation result of the individual crop and the confidence results corresponding to the various postures of the leaves; wherein the postures at least include normal leaves, withered leaves, and curled leaves, and the corresponding postures are output according to the confidence results.
[0013] As some preferred solutions, the first recognition model is a neural network model.
[0014] Another object of the present invention is to disclose a system for identifying the state of agricultural crops based on image analysis, which is used to implement a signed method for identifying the state of agricultural crops based on image analysis, and the system is applied to greenhouse fruit and vegetable cultivation; the system includes: a first processor, which is arranged in the drone and is used to execute S100-S740 and S711-S12; a first communication module, which is used to send the second determination result and posture generated in the first processor to a remote end; a second processor, which is arranged in the greenhouse, and the second processor is at least connected to and controls a temperature regulator and a humidity regulator; a second communication module, which is used to receive the second determination result sent by the first communication module, and transmit the second determination result to the second processor.
[0015] The beneficial effects of the present invention are:
[0016] The method disclosed in the present invention obtains the state of agricultural crops by means of color recognition, obtains corresponding images by two to three rounds of flying of the drone, and processes the color information of the images; compared with the prior art method of identifying the state of agricultural crops by means of neural network models, the present invention simplifies the state recognition of agricultural crops to color recognition, reduces the required data processing amount, and thus improves the recognition efficiency.
[0017] Moreover, since the status of agricultural crops is obtained based on color processing, identification is less affected by factors such as region, climate, crop type, and growth stage. Accurate identification can be performed by simply adding corresponding conditions.
[0018] In addition, through multiple rounds of flights and multi-angle image acquisition, the method of the present invention can obtain more comprehensive information on the status of agricultural crops. For example, the first high-altitude flight can find abnormal areas; the second low-altitude flight can obtain more accurate regional color information, thereby more accurately judging the status of crops.
[0019] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0021] Figure 1 A flow chart of a method for identifying the state of agricultural crops based on image molecules according to an embodiment of the present invention is shown;
[0022] Figure 2 A flow chart of a method for identifying the state of agricultural crops based on image molecules according to an embodiment of the second aspect of the present invention is shown;
[0023] Figure 3 A flow chart of a method for identifying the state of agricultural crops based on image molecules according to an embodiment of the third aspect of the present invention is shown. DETAILED DESCRIPTION
[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0025] In the prior art, the identification of crops is mostly done by taking pictures with drones, and then several crops in the pictures are identified separately through algorithms, and finally the status of each crop is identified according to the pre-trained model. Alternatively, based on satellite images, the spectrum is analyzed to determine whether the crops in the target area are normal. The above two methods are currently the main means of identifying the status of crops based on images. In fact, when acquiring images of large areas of crops, the color information in the images obtained by overhead shooting and the leaf color of a single crop can well reflect its growth status. To this end, the present embodiment discloses a method for identifying the status of crops based on image analysis. The method of this embodiment requires the use of drones to collect images. The drone can provide accurate positioning in complex terrain and is equipped with a high-resolution camera.
[0026] refer to Figure 1 Understanding the method in this embodiment specifically includes the following steps:
[0027] S100, obtaining a target image of a target area taken by a drone, wherein at least a portion of the target area is planted with crops; in different examples, the crops may be different crops, and the drone obtains the target image by taking a bird's-eye view of the target area during flight.
[0028] It is understandable that crops in different seasons have different colors. The target image obtained by aerial photography is a color image. Identifying the color of the target image can determine whether the crops in the target area are normal. Please refer to the subsequent steps for details.
[0029] S200, obtaining pixel information of the target image as first pixel information. In this example, the first pixel information includes color information. Specifically, a multi-spectral camera can be used to obtain the target image in the air to obtain multi-band reflectance information of each pixel point of the target image. The matrix of the first pixel information is defined as I. The matrix I contains information of each pixel point. For example, one pixel point is I (x,y) ;
[0030] The drone is equipped with a multispectral camera, and the acquired multispectral image can be divided into multiple single-band images, each of which records the reflectivity value of a specific band. For example, if there are N bands:
[0031] I (x,y) =[R 1(x,y) , R 2(x,y) , ..., R N(x,y) ]
[0032] Among them, R i(x,y) Represents pixel I (x,y)The reflectance value in the i-th band, where i∈(1,N);
[0033] More specifically, the reflectance value of each band can be calculated using the following formula:
[0034]
[0035] Among them, L i(x,y) is the measured value of the pixel point (x, y) in the i-th band; L i,dark is the dark reference value in the ith band, the dark reference value is a preset parameter, L i,white is the white reference value in the ith band, and the white reference value is a preset parameter.
[0036] The first pixel information is input into the classification recognition model to obtain a classification result, and the classification result includes a normal result and an abnormal result; specifically, the binary classification model pre-trained by the classification model outputs the normal color part as a normal result, and the abnormal part as an abnormal result. For example, taking fruits and vegetables grown in summer as an example, the green part in the first pixel is classified as the normal result, and the non-green part is classified as an abnormal result. The specific division process of the classification recognition model will be explained in detail in subsequent specific examples.
[0037] S300, generating a mask according to the classification result, operating and processing the mask to divide each color block in the target image, and the color block is at least divided into a normal block and an abnormal block; it can be understood that dividing the target image into different color blocks can be used for: the drone locates the crops that may be abnormal during the second round of flight, specifically, the drone will fly above the crops corresponding to the abnormal blocks to collect the second round of images. The specific process of the mask processing is described in the subsequent examples.
[0038] Based on the foregoing, it can be understood that the area division in this embodiment is based on color division, which is different from the prior art in which block division is obtained by feature recognition method. Color recognition can reduce the parameters that need to be processed and avoid the influence of different environments and regions on the judgment accuracy.
[0039] S400, in combination with the above steps, the drone can obtain a preliminary target image by taking a bird's-eye view of the crops, but the target image may have insufficient clarity or resolution due to the distance between the drone and the crops. Therefore, in this step, the position of the abnormal block in the target area is located as a sub-target area; it is understandable that locating the sub-target area can further determine whether there is an abnormal crop in the abnormal block initially screened out.
[0040] S500, obtaining a sub-target image of the sub-target area photographed by the drone;
[0041] S600, obtaining pixel information containing color of the sub-target image as second pixel information, and comparing the second pixel information with a preset color range to obtain a color determination result. In this step, the specific data storage method and expression form of the second pixel information including color information are basically the same as those of the first pixel information, and will not be described in detail here. The color range is a plurality of preset thresholds, and different color determination results are obtained according to the threshold determination.
[0042] It should be understood that the first pixel information is classified and identified by the classification recognition model, the purpose of which is to divide the normal block from the abnormal block in the subsequent steps, and the color recognition of the second pixel information is to output the final result, that is, to output the result of the sub-target area after recognition. The specific process can be understood by referring to the subsequent specific examples.
[0043] In the foregoing example, further reference is made to Figure 2 To understand, the drone actually performs two rounds of image acquisition, specifically including:
[0044] In S100, the drone performs a first flight to acquire the target image. During the first flight, the drone records the coordinate information of the target area and projects the coordinate information into the target image.
[0045] In S500, the drone performs a second flight to acquire the sub-target image. When performing the second flight, the drone reads the coordinate information corresponding to the sub-target area for: acquiring the sub-target image within the sub-target area. It can be understood that the drone flies to the sub-target area corresponding to the abnormal block according to the coordinate information to acquire the image. The altitude at which the drone performs the first flight is higher than the altitude of the second flight.
[0046] It is understandable that in this embodiment, in order to improve the recognition efficiency of agricultural crops, the target area is photographed from a higher position during the first recognition process to quickly screen out the abnormal blocks where the crop status may be abnormal, and then the abnormal blocks are used as sub-target areas for a second round of photography to obtain the specific sub-target image. The sub-target image is compared with the color range to obtain the color determination result. It is understandable that the color determination result can reflect the growth status of the agricultural crops in the sub-target area.
[0047] In some specific examples, S610 executed after S600 is also included;
[0048] S610, the color judgment result includes at least normal color and abnormal color; the color judgment result is obtained, and the proportion of abnormal colors in the color judgment result is calculated as a first ratio value. It can be understood that the ratio of abnormal colors can represent whether the agricultural crops in the corresponding sub-target area are indeed abnormal, that is, the result of whether the crops are abnormal can be obtained according to the first ratio value. Specifically, the first ratio value is compared with the first preset threshold value and then the first judgment result is output. The first judgment result includes normal crops and abnormal crops. The abnormal crops can be sent to a remote mobile device, and the user determines whether the next step is required based on the abnormal crop image received by the mobile device.
[0049] First aspect example:
[0050] In combination with the above content, in order to better illustrate the method of the present invention, this example will illustrate the method of this embodiment with a specific example. In the example, the crops are fruits and vegetables, the single crop target is the leaves of fruits and vegetables, and the time is summer.
[0051] As mentioned above, when the drone performs the first flight again, it takes a bird's-eye view of the target area at a high altitude. Due to the long distance, the color accuracy may be poor when acquiring the target image. In order to avoid misclassifying an abnormal result close to a normal result as a normal result, a simple threshold determination cannot be used to identify the first pixel information. For example, in this example, the normal result should be green, and using a threshold determination may misclassify dark green as green.
[0052] Continuing with the pixel I mentioned in the previous example (x,y) For example, the spectral reflectance vector of each pixel is input into the classification recognition model to obtain the probability P that the pixel belongs to the non-green category (i.e., abnormal result). In this example, the classification recognition model is obtained based on training of several overhead images, different training sets can be used for different agricultural crops, and the classification threshold T of the model is set, and the classification decision is made according to the output probability P.
[0053] In this example, it is necessary to avoid misclassifying abnormal results as normal results, so the classification threshold T needs to be set to a value not less than 0.7.
[0054] Then a mask operation is performed according to the classification result Class (x,y)Generate a mask matrix M. Specifically, convert the multi-band reflectance information into color information, and use a common color space (such as RGB or HSV) for color recognition. More specifically, convert the RGB value into the HSV value for recognition. Finally, set different masks for different color results, and then apply the mask matrix M to the original target image to obtain the color block division result.
[0055] Second aspect embodiment;
[0056] Based on the first embodiment, the state recognition of fruits and vegetables is continued as an example, and combined with Figure 2 It is understood that the method for identifying the state of agricultural crops based on image analysis in this example also includes S710-S720 performed after S610;
[0057] S710, the drone performs a third flight; in this example, during the third flight, the drone hovers over the sub-target area and further moves down to be close to the leaves of fruits and vegetables. For example, a support for supporting fruits and vegetables is provided in the area where the fruits and vegetables are planted, and the drone hovers next to the support to perform a third shot.
[0058] During shooting, the drone adjusts the angle of the camera or the overall posture, and obtains the image of any single agricultural crop target in the abnormal block as the first individual image; wherein the acquired single agricultural crop target is acquired randomly, and the single agricultural crop target is the leaves of fruits and vegetables, and the first individual image is a front view of the leaves.
[0059] S720, obtaining pixel information containing color of the first individual image as third pixel information, and comparing the third pixel information with the preset color range to obtain color distribution. Specifically, the drone randomly selects several leaves in the sub-target area to shoot, and obtains several front images of the leaves as the first individual image.
[0060] In this step, the drone takes a frontal shot of the blade to obtain a high-resolution image; the image is processed to obtain the third pixel information, wherein the color information can be expressed in an RGB color space or an HSV color space. The third pixel information is compared with a preset color range to obtain a color distribution.
[0061] In some preferred examples, S730 to S750 are further included after S720;
[0062] S730, obtaining the color distribution, the color distribution at least includes the colors distributed in the first individual image, specifically, it may include a first color and a second color, and in different examples, it may also include multiple colors, for example, the first color is green, the second color is dark green, and a third color is also included, and the third color is other colors. Calculate the proportion of the first color in the entire leaf as a second ratio value, compare the second ratio value with the second preset threshold value and output a second judgment result, the second judgment result includes a single crop being normal and a single crop being abnormal. For example, in the first individual image of a leaf, the total number of pixels is 1000, of which the number of pixels that meet the green range is 800, and the preset threshold is 0.85; then the second ratio value is preset to 0.8; therefore, it is determined that the leaf has an abnormal color.
[0063] Since in S730, a single crop is obtained randomly, the corresponding confidence interval can be calculated by the normal number of the single crop, as shown in S740.
[0064] S740: According to the second determination result, record the number of results that the single crop is normal, and calculate the confidence interval that the single crop is normal;
[0065] In a specific example, assume that m individual crops are randomly selected in the sub-target area for photographing and analysis, and n individual crops are determined to be normal. According to the binomial distribution, the confidence interval can be calculated:
[0066]
[0067] in, is the normal proportion of a single crop; at a 95% confidence level, using the standard error estimate, the confidence interval can be expressed as:
[0068]
[0069] Among them, Z α / 2 is the critical value corresponding to the standard normal distribution. At the 95% confidence level, Z α / 2 ≈1.96.
[0070] For example, if 20 leaves are photographed in the sub-target area and 15 of them are judged to be normal, the proportion of normal leaves in a single crop is is 0.75.
[0071] The 95% confidence interval is:
[0072]
[0073] S750, comparing the first ratio value with the confidence interval; it is understood that if the first ratio value falls within the confidence interval, it indicates that the first determination result has a high degree of credibility; if it does not meet the confidence interval, it indicates that the first determination result is unreliable, and the first preset threshold needs to be readjusted.
[0074] When the first ratio value falls within the confidence interval, outputting the first determination result and / or the second determination result;
[0075] When the first ratio value does not fall within the confidence interval, the first preset threshold is increased by the first increase value and S610 is performed again. It can be understood that increasing the preset threshold can make the classification model more strictly determine the abnormal area, and as the number of detections increases, by gradually adjusting the preset threshold, the performance of the classification recognition model can be gradually optimized. In some cases, the first increase value can be set according to specific circumstances. It can be understood that by updating the first preset threshold, it is possible to adapt to different agricultural crop environments. Compared with the prior art method of using a trained model to identify agricultural crops, this example can adjust the first preset threshold so that the model can be adaptively adjusted according to the specific scenario, thereby further improving the adaptability to different environments.
[0076] In some specific examples, the second determination result includes: a first sub-result, the first sub-result includes a normal blade and an abnormal blade. The first sub-result is used to match the first determination result. When the first sub-result is consistent with the first determination result, it means that the first determination result is reliable. And a second sub-result, the second sub-result is used to output the specific state of the blade. In this example, the second sub-result includes at least green blade, yellow blade, dark green blade, and white blade.
[0077] The third aspect embodiment;
[0078] Based on the second aspect embodiment, this example combines Figure 3 It is understood that the method further includes S711 executed synchronously with S710;
[0079] S711, the drone performs a third flight to obtain a second individual image of any single crop target in the abnormal block; as mentioned above, the single crop is a fruit or vegetable, and the single crop target is a leaf of the fruit or vegetable. In combination with the above, the first individual image is a front image of the leaf, and the second individual image in this example is a side image of the leaf.
[0080] The state of the leaves can reflect the state of the crop. Compared with the first embodiment, this example obtains the first individual image and the second individual image to determine the state of the leaves. It is understandable that when the drone performs the third flight, it randomly obtains multiple leaves for individual identification, so as to accurately obtain the state of each leaf and then determine the state of the crop. The processing of the first individual image has been explained in the previous embodiment and will not be repeated here. The processing of the second individual image will be described in detail below.
[0081] Continue to combine Figure 3 Explain S712 executed after S711;
[0082] S712: Input the second individual image into the first recognition model to obtain the confidence results corresponding to the radius estimation results of the individual crop and the various postures of the leaves, and output the corresponding posture according to the confidence results.
[0083] Specifically, the first recognition model is a neural network model; the neural network model's recognition process for leaf posture includes: inputting the sub-target image into the input layer of the neural network model, and extracting feature information of the sub-target image through a convolutional layer, a pooling layer, and a fully connected layer; after the fully connected layer, classifying the extracted features through a softmax layer, and outputting confidence results for the posture categories of each leaf of each sub-target; determining the final posture category of each pressed tablet based on the confidence results output by the softmax layer, and taking the category with the highest confidence result as the corresponding posture; finally, outputting the posture of each sub-target, the postures including normal leaves, withered leaves, and curled leaves.
[0084] The obtained leaf posture and color can be combined to obtain the accurate state of the crop, and the specific process can be understood by referring to the fourth embodiment.
[0085] Fourth aspect embodiment;
[0086] In combination with the above methods, this embodiment discloses a crop state recognition system based on image analysis, which is used to implement the crop state recognition method based on image analysis described in any one of the above embodiments, and apply it to greenhouse fruit and vegetable planting; the system in this example includes:
[0087] A first processor, the first processor is arranged in the UAV, and is used to execute S100-S740 and S711-S12. The first communication module is used to: send the second determination result generated in the first processor and the posture of a single agricultural crop to a remote end; a second processor, the second processor is arranged in the greenhouse, and the second processor is at least connected to and controls a temperature regulator and a humidity regulator; a second communication module, the second communication module is used to: receive the second determination result sent by the first communication module, and transmit the second determination result to the second processor.
[0088] In a specific example, the first processor uses an embedded processor, such as NVIDIA JetsonTX2 or NVIDIA Jetson Xavier NX, and has a built-in GPU acceleration module, which connects the camera and other sensors through a USB 3.0 or PCIe interface to ensure fast data transmission and processing. The second processor is an industrial-grade computer. The first communication module is a wireless communication module built into the drone, such as a LoRa, ZigBee or Wi-Fi module, which is used to transmit data in real time during the flight of the drone. The second communication module is a wireless communication module set in a greenhouse, such as a LoRa gateway or a Wi-Fi access point, which is used to receive determination results and image data from the drone.
[0089] In a specific case:
[0090] The second processor performs the following determination based on the second determination result received by the second processor and the result of the posture of the single crop:
[0091] The second sub-result is received, and if the specific state of the leaf is green, it is determined to be normal, if the specific state of the leaf is dark green, it is determined that the fertilizer supply is too much, and the result is sent to the user's mobile device. If the specific state of the leaf is yellow, it is determined that the fertilizer supply is insufficient, and the result is sent to the user's mobile device.
[0092] And the second processor controls the equipment in the greenhouse to adjust, specifically: the second processor fuses the results of the posture and color of a single agricultural crop;
[0093] When the posture of a single crop is drooping and the specific state of the leaves is yellow, it is determined to be water-deficient, and the second processor controls the sprinkler system to automatically irrigate. When the posture of a single crop is curling upward, the second processor reads preset data, which includes: fruits and vegetables are in the seedling stage and flowering and fruiting stage or in the pre-bud stage.
[0094] If the plant is in the seedling stage and the flowering and fruiting stage, the edge of the leaf is identified, wherein the color distribution identification has been performed in S730, so the second processing can identify the color of the leaf edge. When the color of the leaf edge is identified as yellow or white, it is determined that the growth temperature is too low, and the second processor controls the temperature regulator to automatically adjust the temperature.
[0095] If it is in the early bud stage, when the color of the leaf edge is identified as yellow or white, it is determined that the growth temperature is too high, and the second processor controls the temperature regulator to automatically adjust and cool down.
[0096] It can be seen that the crop state recognition system based on image analysis can adjust the fruit and vegetable growing environment in the greenhouse based on the recognition structure, save manpower, and is particularly suitable for automated agricultural planting.
[0097] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying crop status based on image analysis, characterized in that: The steps include: S100, obtaining a target image taken by a drone of a target area, wherein at least a portion of the target area is planted with crops, and the target image is a color image; S200, obtaining pixel information of the target image as first pixel information, and inputting the first pixel information into a classification recognition model to obtain a classification result, wherein the classification result includes a normal result and an abnormal result; S300, generating a mask according to the classification result, operating and processing the mask to divide each color block in the target image, wherein the color block is at least divided into a normal block and an abnormal block; S400, locating the position of the abnormal block in the target area as a sub-target area; S500, obtaining a sub-target image of the sub-target area photographed by the drone; S600: Obtain pixel information containing color of the sub-target image as second pixel information, and compare the second pixel information with a preset color range to obtain a color determination result.
2. The method for identifying crop status based on image analysis according to claim 1, characterized in that: In S100, the drone performs a first flight to acquire the target image, and during the first flight, records coordinate information of the target area and projects the coordinate information into the target image; In S500, the drone performs a second flight to acquire the sub-target image. When performing the second flight, the drone reads the coordinate information corresponding to the sub-target area to: acquire the sub-target image within the range of the sub-target area; The altitude at which the UAV performs the first flight is higher than the altitude at which the UAV performs the second flight.
3. The method for identifying crop status based on image analysis according to claim 2, characterized in that: Also included is S610 executed after S600; S610, the color judgment result includes at least normal color and abnormal color; the color judgment result is obtained, the proportion of abnormal color in the color judgment result is calculated as a first proportion value, and the first proportion value is compared with a first preset threshold value to output a first judgment result, and the first judgment result includes normal crop and abnormal crop.
4. The method for identifying crop status based on image analysis according to claim 3, characterized in that: It also includes S710-S720 performed after S610; S710, the UAV performs a third flight; When performing the third flight, an image of any single agricultural crop target in the abnormal block is acquired as a first body image; S720: Obtain pixel information containing color of the first individual image as third pixel information, and compare the third pixel information with the preset color range to obtain color distribution.
5. The method for identifying crop status based on image analysis according to claim 4, characterized in that: It also includes S730-S750 performed after S720; S730, obtaining the color distribution, the color distribution at least including a first color and a second color distributed in the first individual image, calculating a ratio of the first color to the second color as a second ratio value, and outputting a second determination result after comparing the second ratio value with a second preset threshold, the second determination result including whether a single crop is normal or abnormal; S740: According to the second determination result, record the number of results that the single crop is normal, and calculate the confidence interval that the single crop is normal; S750, comparing the first ratio value with the confidence interval; When the first ratio value falls within the confidence interval, outputting the first determination result and / or the second determination result; When the first ratio value does not fall within the confidence interval, the first preset threshold is increased by a first increase value and S610 is performed again.
6. The method for identifying crop status based on image analysis according to claim 5, characterized in that: The agricultural crops are fruits and vegetables, the single agricultural crop target is a leaf of the fruit and vegetable, and the first individual image is a front view of the leaf; The second determination result includes: The first sub-result includes normal leaves and abnormal leaves; The second sub-fruits include green leaves, yellow leaves, dark green leaves, and white leaves.
7. The method for identifying crop status based on image analysis according to claim 6, characterized in that: Also included is S711 executed synchronously with S710; S711, when the UAV performs the third flight, obtaining a second individual image of any single agricultural crop target in the abnormal block; The second individual image is an image of the side surface of the blade.
8. The method for identifying crop status based on image analysis according to claim 7, characterized in that: Also included is S712 executed after S711; S712, inputting the second individual image into the first recognition model to obtain the radius estimation result of the individual crop and the confidence results corresponding to the multiple postures of the leaves; The posture at least includes normal leaves, withered leaves, and curled leaves, and the corresponding posture is output according to the confidence result.
9. The method for identifying crop status based on image analysis according to claim 8, characterized in that: The first recognition model is a neural network model.
10. A system for identifying the state of agricultural crops based on image analysis, used to implement the method for identifying the state of agricultural crops based on image analysis according to claim 8, characterized in that: Applied to greenhouse fruit and vegetable cultivation; The system includes: A first processor, which is disposed in the drone and is used to execute S100-S740 and S711-S12; A first communication module, the first communication module is used to: send the second determination result and posture generated in the first processor to a remote end; A second processor, the second processor is arranged in the greenhouse, and the second processor is at least connected to and controls a temperature regulator and a humidity regulator; The second communication module is used for receiving the second determination result sent by the first communication module, and transmitting the second determination result to the second processor.
Citation Information
Patent Citations
Bottle culture plant planting control method
CN104705080A
Method and system for intelligent identification and accurate pesticide spraying using unmanned aerial vehicles
CN106585992A
Pest diagnosis system of agricultural greenhouse
CN109406517A
Crop recognition method, device, equipment and storage medium
CN113850132A
Bursaphelenchus xylophilus disease detection and analysis system based on big data
CN116994159A
Cited By
Greening plant withering detection method and system based on dynamic patrol
CN121527547A
Greening plant withering detection method and system based on dynamic patrol
CN121527547B
