Harvester operation quality evaluation method and system based on grain state

Abstract

The invention provides a grain state-based harvester operation quality evaluation method and system, and relates to the technical field of harvester operation quality evaluation.Historical grain images in a harvester storage bin are acquired to form an image data set, and the images acquired in real time are preprocessed to obtain an image data set; the method comprises the following steps: inputting into a trained grain state recognition model, outputting a first label, calculating a crushing rate and a crushing degree, preprocessing an image collected in real time, inputting into a trained impurity recognition model, outputting a second label, calculating an impurity content, obtaining historical environment data and material characteristics of grain straws, and calculating an environment interference coefficient. The grain distribution uniformity is calculated according to the breakage rate and the breakage degree of the grains and the coordinate position of image acquisition, and the operation quality is judged according to the environment interference coefficients and the impurity rates at different time intervals and according to the grain distribution uniformity and the breakage rate and the breakage degree at different time intervals.

Description

Technical Field

[0001] The present invention relates to the technical field of harvester operation quality assessment, and in particular to a harvester operation quality assessment method and system based on grain status. Background Art

[0002] In the past few decades, harvesters have been the core operating equipment in agricultural production, and their operating quality has directly affected the yield and quality of grain. However, the traditional method of evaluating the operating quality of harvesters mainly relies on manual visual inspection and empirical judgment, lacking precise quantitative standards, resulting in a highly subjective and inefficient evaluation process, especially in large-scale operations. This traditional evaluation method cannot fully meet the needs of modern agriculture for efficient, accurate, and automated operating quality monitoring.

[0003] With the rapid development of smart agriculture, methods based on machine vision and data analysis have gradually been applied. These methods can evaluate the operation quality of harvesters in real time by collecting and analyzing operation data in real time. For example, by collecting images in the storage bin of the harvester, combined with image processing and machine learning technology, key information such as grain damage, impurity content, and grain distribution can be accurately identified, thereby realizing automatic evaluation of operation quality. In addition, the impact of environmental factors on operation quality during the operation process has also received increasing attention, and traditional technologies often ignore the role of environmental interference.

[0004] Therefore, how to combine the harvester's real-time operation data, environmental change information and advanced intelligent recognition technology to accurately and comprehensively evaluate the harvester's operation quality has become an important issue to solve the shortcomings of existing technologies. This requires not only real-time monitoring of the operation process, but also data analysis, model training, image recognition and other technologies to cope with the complex and changing agricultural operation environment and high-precision evaluation needs.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The object of the present invention is to provide a method and system for evaluating the operation quality of a harvester based on grain status, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for evaluating harvester operation quality based on grain status, comprising the following specific steps:

[0009] Step 1: Obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval, preprocess the images to form an image data set, and establish a coordinate system by recording the harvesting trajectory when harvesting grains and making a one-to-one correspondence between the harvester's position at the time of image acquisition and the acquired image;

[0010] Step 2: The kernels in the image data set are labeled with a first label according to the kernel breakage state and severity, a kernel state recognition model is established based on a convolutional neural network, and the model is trained. After preprocessing the real-time collected images, the images are input into the trained kernel state recognition model, the first label is output, and the breakage rate and degree of breakage are calculated;

[0011] Step 3: Mark the kernels in the image data set with a second label according to the number and size of impurities contained, establish an impurity recognition model based on a convolutional neural network, and train it. Preprocess the real-time collected images, input them into the trained impurity recognition model, output the second label, and calculate the impurity rate.

[0012] Step 4: Obtain historical environmental data and material properties of grain straw, discretize the historical environmental data according to the time interval between photos, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, take the grain breakage rate and degree of breakage, and the adjacent values ​​of the coordinate position at the time of image acquisition as features, classify the grain distribution, and calculate the uniformity of grain distribution based on the classification results;

[0013] Step 5: Calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals. Calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals. Judge the operation quality based on the anti-interference coefficient and grain harvesting quality.

[0014] Furthermore, the method for preprocessing the image is image normalization, and the specific calculation formula is:

[0015] The method for preprocessing the image is to normalize the pixels of the image. The calculation formula is:

[0016]

[0017] Among them, Xsg is the normalized pixel value, and Xs is the pixel value of the image.

[0018] Furthermore, the convolutional neural network specifically includes a convolution layer, an activation layer, a pooling layer and an output layer;

[0019] Convolutional layer: extracts features from the bounding box and label of the input image through convolution operation, and the calculation formula is:

[0020]

[0021] Where I is the input image matrix, K is the convolution kernel, (xc, yc) is the coordinate of the output matrix, and K(ic, jc) is the value of the convolution kernel in the icth row and jcth column;

[0022] The activation layer is calculated as:

[0023] ReLU(xz)=max(0,xz)

[0024] Among them, xz is the coordinate of the output matrix of the convolutional layer output;

[0025] The calculation formula of the pooling layer is:

[0026] MaxPooling(xd,yd)=max{I(xc+ic,yc+jc)|ic,jc∈(3,3)}

[0027] Where I is the input matrix, MaxPooling(xd,yd) is the maximum pooling output;

[0028] The calculation formula of the output layer is:

[0029] y=W*xm+b

[0030] Among them, xm is the output of the pooling layer, y is the result of image recognition, W is the weight, and b is the bias parameter;

[0031] The image dataset is used as the input of the grain state recognition model, the first label is used as the output of the grain state recognition model, and the grain state recognition model is trained.

[0032] Furthermore, the grain crushing state is divided into epidermal crushing and whole crushing;

[0033] The severity levels are classified into slight epidermal fragmentation, severe epidermal fragmentation, slight overall fragmentation, and complete overall fragmentation;

[0034] The label content includes the state, severity and number of kernels broken;

[0035] The calculation formula for calculating the crushing rate is:

[0036]

[0037] Among them, Nbp is the number of kernels with broken skin, Nzp is the number of kernels with broken whole, Nzl is the number of all kernels, and Lps is the breakage rate;

[0038] The calculation formula for calculating the degree of fragmentation is:

[0039]

[0040] Among them, Pps is the degree of breakage, Nbp1 and Nbp2 are the number of kernels with slightly broken skin and severely broken skin, respectively, Nzp1 and Nzp2 are the number of kernels with slightly broken skin and completely broken skin, respectively, α1 is the influence coefficient of kernels with broken skin, and α2 is the influence coefficient of kernels with completely broken skin.

[0041] Furthermore, the training steps of the impurity recognition model are as follows: the image data set is used as the input of the impurity recognition model, and the second label is used as the output of the impurity recognition model to train the impurity recognition model.

[0042] The calculation formula of the impurity content is:

[0043]

[0044] Among them, Phz is the impurity rate, Sxs is the pixel area of ​​the image, and Szzia is the pixel area of ​​the iath impurity.

[0045] Further, the environmental data includes temperature value, humidity value and wind speed value;

[0046] The calculation formula for calculating the environmental interference coefficient is:

[0047]

[0048] Among them, Phg is the environmental interference coefficient, Wsd is the humidity value, Twd is the temperature value, ka is the correction coefficient, Vfs is the wind speed value, Pcz is the material characteristic of the straw, and t is the current time interval;

[0049] The calculation formula of the adjacent value of the coordinate position is:

[0050]

[0051] Among them, PzbDlj is the adjacent value of the grain coordinate position, (x iq ,y iq ), (x ip ,y ip ) are the coordinate positions of the iqth and ipth seeds respectively.

[0052] Furthermore, the specific steps of classifying the grain distribution are:

[0053] Select the fragmentation rate Lps, fragmentation degree Pps, and proximity value Pzb as the feature vector of the coordinate position corresponding to each image recognition moment;

[0054] Randomly select K data nodes as the initial centroids, where K represents the number of categories after clustering and K is a positive integer. The i-th initial centroid is expressed as: C i [Lps(i), Pps(i), Pzb(i)], Lps(i), Pps(i) and Pzb(i) respectively represent the fragmentation rate, fragmentation degree and proximity value of the i-th data node as the initial centroid, i is a positive integer, and i = 1, 2, ..., K;

[0055] For each feature vector at each coordinate position, calculate its distance to each initial centroid and assign it to the cluster represented by the nearest centroid. The formula for calculating the distance to the initial centroid is:

[0056]

[0057] Where d(i, j) represents the distance between the i-th initial centroid and the j-th data node, Lps(j), Pps(j) and Pzb(j) represent the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the j-th data node, respectively;

[0058] For each cluster, after each clustering is completed, the mean of all points in the cluster is recalculated, and this mean is used as the feature data of the new centroid. The update formula of the i-th centroid feature data is:

[0059]

[0060] Where m represents the number of data nodes assigned to the i-th centroid, Lps(i) old 、Pps(i) old and Pzb(i) old Indicates the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the data node assigned to the i-th centroid, Lps(i) new 、Pps(i) new and Pzb(i) new Indicates the fragmentation rate, fragmentation degree and proximity value of the updated centroid;

[0061] Based on the updated centroid feature vector, clustering is performed again until the change in the position of all centroids is less than the threshold, then the clustering is considered stable and the clustering is terminated;

[0062] The calculation formula for calculating the uniformity of grain distribution according to the classification results is:

[0063]

[0064] Among them, Pdx is the uniformity of grain distribution, nz ik is the number of coordinate points in the ikth cluster, and K is the number of clusters.

[0065] Furthermore, the calculation formula for calculating the anti-interference coefficient of the harvester is:

[0066]

[0067] Among them, Gkg is the anti-interference coefficient, Phg is the environmental interference coefficient, Phz is the impurity rate, ic is the image collected at the icth time, and nc is the number of image collections;

[0068] The calculation formula for calculating the grain harvest quality is:

[0069]

[0070] Among them, Lpsh is the stable value of the breakage rate, Ppsh is the stable value of the breakage degree, Lps is the breakage rate, Pps is the breakage degree, ic is the image collected at the icth time, nc is the number of image collections, Pdx is the uniformity of grain distribution, and Gzz is the grain harvesting quality; the calculation formula for judging the operation quality based on the anti-interference coefficient and the grain harvesting quality is:

[0071] Gl=Gkg*α1+Gzz*α2

[0072] Among them, Gl is the operating quality of the harvester, α1 and α2 are the weights of the anti-interference coefficient and the grain harvesting quality, respectively, and α1+α2=1.

[0073] The present invention also provides a harvester operation quality assessment system based on grain status, wherein the harvester operation quality assessment system is used to execute the above-mentioned harvester operation quality assessment method based on grain status, comprising:

[0074] A data acquisition module, which is used to obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval when the harvester is operating, and preprocess the images to form an image data set, and establish a coordinate system by recording the harvesting trajectory when harvesting grains, and making a one-to-one correspondence between the position of the harvester at the time of image acquisition and the acquired image;

[0075] A grain recognition module, wherein the grain recognition module is used to mark the grains in the image data set with a first label according to the grain breakage state and severity, establish a grain state recognition model based on a convolutional neural network, and train the model, pre-process the real-time collected image, input it into the trained grain state recognition model, output the first label, and calculate the breakage rate and breakage degree;

[0076] An impurity identification module is used to mark the kernels in the image data set with a second label according to the number and size of impurities contained, establish an impurity identification model based on a convolutional neural network, and perform training, pre-process the real-time collected image, input it into the trained impurity identification model, output the second label, and calculate the impurity content;

[0077] An interference analysis module, which is used to obtain historical environmental data and material properties of grain straw, discretize the historical environmental data according to the time interval between photographing, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, classify the grain distribution by taking the grain breakage rate and degree of breakage, and the adjacent values ​​of the coordinate position at the time of image acquisition as features, and calculate the uniformity of grain distribution according to the classification results;

[0078] The quality assessment module is used to calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals, calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals, and judge the operation quality according to the anti-interference coefficient and the grain harvesting quality.

[0079] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention greatly improves the efficiency and accuracy of the harvester operation quality assessment by introducing an automated assessment system based on image recognition. Through the convolutional neural network technology, the broken state and impurities of the grains can be accurately identified and classified, avoiding the subjective errors of manual inspection. The method of specialized processing based on the establishment of grain recognition models and impurity recognition models can effectively reduce the risk of misidentification, and improve the accuracy and reliability of the assessment in complex environments. At the same time, the calculation of the interference coefficient in combination with environmental data and the material characteristics of grain straw provides a more comprehensive basis for evaluating the operation quality. It not only improves the real-time performance of operation quality monitoring, but also effectively guides the adjustment and optimization of the harvester, and improves the overall efficiency of agricultural production.

[0080] The present invention also calculates the anti-interference coefficient of the harvester and the grain harvesting quality by comprehensively analyzing multiple indicators such as the environmental interference coefficient, impurity content, grain distribution uniformity, and degree of damage, and finally judges the operation quality. This technology integrates the quantitative analysis of multiple factors and can comprehensively evaluate the operation performance of the harvester in a complex environment. Through the dynamically adjusted anti-interference coefficient, it can cope with the instability caused by environmental changes during the operation, thereby providing accurate quality assessment in different time periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0082] Figure 2It is a schematic diagram of the overall system structure of the present invention. DETAILED DESCRIPTION

[0083] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0084] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0085] Example:

[0086] See also Figure 1 , the present invention provides a technical solution:

[0087] A method for evaluating harvester operation quality based on grain status, comprising the following specific steps:

[0088] Step 1: Obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval, preprocess the images to form an image dataset, and establish a coordinate system by recording the harvesting trajectory when harvesting grains and making a one-to-one correspondence between the harvester's position at the time of image acquisition and the acquired image.

[0089] By acquiring and preprocessing grain images at the same time interval, and recording the trajectory and position of the harvester, the temporal and spatial consistency of the image data can be ensured. This one-to-one correspondence in time and space makes subsequent data analysis more accurate. For example, being able to accurately match the grain image at a certain moment with the environmental state of the harvester's location at that moment is crucial for each subsequent step of evaluating the quality of the operation.

[0090] Automatically collecting images and associating them with location data greatly improves the efficiency and accuracy of data acquisition. Manual methods are not only easily interfered by human factors, but also cannot obtain large-scale data in real time during the operation. This method can reflect the changes in the entire harvesting process in real time through automated data collection, ensuring the real-time nature of the evaluation.

[0091] Through accurate image acquisition and positioning information, subsequent image analysis based on convolutional neural networks and other technologies can more accurately identify kernel damage, impurities, etc., thereby providing more scientific data support for overall quality assessment. This information is not only helpful for quality assessment of a single operation point, but also can achieve intelligent decision-making and optimization in the overall operation process, further improving operation quality and production efficiency.

[0092] The steps to establish a coordinate system are to select a fixed geographic reference point as the origin of the coordinate system. This reference point can be the starting position of the harvester (for example, the position at the beginning of the operation) or a known position related to the harvester's path. Determine the coordinate axes of the coordinate system, where one axis (such as the X-axis) represents the horizontal direction (east-west direction) and the other axis (such as the Y-axis) represents the vertical direction (north-south direction). The harvester's GPS system records the position coordinates at each moment. For example, at time T1, the GPS position of the harvester is (X1, Y1); at time T2, the GPS position of the harvester is (X2, Y2). Each time a grain image is collected, the timestamp of the image collection is recorded (for example, the T1 image is associated with the T1 timestamp, and the T2 image is associated with the T2 timestamp). Create a data table to save the timestamp and corresponding coordinates of each image (for example, image 1 corresponds to the coordinates (X1, Y1), and image 2 corresponds to the coordinates (X2, Y2)).

[0093] In this embodiment, the method for preprocessing the image is image normalization, and the specific calculation formula is:

[0094] The method for preprocessing the image is to normalize the pixels of the image. The calculation formula is:

[0095]

[0096] Among them, Xsg is the normalized pixel value, and Xs is the pixel value of the image.

[0097] Image normalization is to adjust the pixel values ​​of an image to a specific range (for example, between 0 and 1) to make the scale of the image data uniform. This is usually achieved through linear transformation to ensure that the numerical range of the image data will not affect the performance of the model due to differences in input data during training. Through image normalization, the numerical range of all input data can be made close, helping the optimization algorithm to converge more quickly and stably, and improving the training efficiency and performance of the model.

[0098] By preprocessing the images to form a standardized data set, it is convenient for the subsequent efficient training and analysis of the images using machine learning or deep learning techniques. The uniformity and structure of this data set provide high-quality input for subsequent steps (such as kernel breakage status recognition, impurity detection, etc.), thereby improving the accuracy and reliability of the model.

[0099] Step 2: Mark the grains in the image dataset with a first label according to the grain breakage status and severity, establish a grain status recognition model based on a convolutional neural network, and train the model. After preprocessing the real-time collected images, input them into the trained grain status recognition model, output the first label, and calculate the breakage rate and degree of breakage.

[0100] Automated grain state recognition through convolutional neural networks can automatically extract effective features from a large number of real-time images for efficient and accurate classification. Convolutional neural networks can effectively identify the degree and type of grain breakage, and can perform identification relatively stably even under different environmental and lighting conditions. Through the training of convolutional neural networks, the broken state of grains (such as slight breakage, severe breakage, etc.) can be carefully classified and marked as the first label. This makes the assessment of the broken state not only more accurate, but also quantifiable to the breakage rate and degree of breakage. This detailed evaluation method can provide more objective and accurate data support for harvester operations, helping operators to adjust operation strategies in real time and optimize harvesting results.

[0101] In this embodiment, the convolutional neural network specifically includes a convolution layer, an activation layer, a pooling layer and an output layer;

[0102] Convolutional layer: extracts features from the bounding box and label of the input image through convolution operation, and the calculation formula is:

[0103]

[0104] Where I is the input image matrix, K is the convolution kernel, (xc, yc) is the coordinate of the output matrix, and K(ic, jc) is the value of the convolution kernel in the icth row and jcth column;

[0105] The activation layer is calculated as:

[0106] ReLU(xz)=max(0,xz)

[0107] Among them, xz is the coordinate of the output matrix of the convolutional layer output;

[0108] The calculation formula of the pooling layer is:

[0109] MaxPooling(xd,yd)=max{I(xc+ic,yc+jc)|ic,jc∈(3,3)}

[0110] Where I is the input matrix, MaxPooling(xd,yd) is the maximum pooling output;

[0111] The calculation formula of the output layer is:

[0112] y=W*xm+b

[0113] Among them, xm is the output of the pooling layer, y is the result of image recognition, W is the weight, and b is the bias parameter;

[0114] The image dataset is used as the input of the grain state recognition model, the first label is used as the output of the grain state recognition model, and the grain state recognition model is trained.

[0115] In this embodiment, the seed crushing state is divided into epidermal crushing and whole crushing;

[0116] The severity levels are classified into slight epidermal fragmentation, severe epidermal fragmentation, slight overall fragmentation, and complete overall fragmentation;

[0117] The label content includes the state, severity and number of kernels broken;

[0118] The calculation formula for calculating the crushing rate is:

[0119]

[0120] Among them, Nbp is the number of kernels with broken skin, Nzp is the number of kernels with broken whole, Nzl is the number of all kernels, and Lps is the breakage rate;

[0121] The calculation formula for calculating the degree of fragmentation is:

[0122]

[0123] Among them, Pps is the degree of breakage, Nbp1 and Nbp2 are the number of kernels with slightly broken skin and severely broken skin, respectively, Nzp1 and Nzp2 are the number of kernels with slightly broken skin and completely broken skin, respectively, α1 is the influence coefficient of kernels with broken skin, and α2 is the influence coefficient of kernels with completely broken skin.

[0124] The breakage rate indicates the proportion of crop grains that are damaged or broken during the harvesting process, usually referring to the ratio of the number of broken grains to the total number of grains. This value can reflect the efficiency of the harvester during operation. The lower the breakage rate, the less damage the harvester causes to the grains during operation, and the better the operation quality. Generally, the lower the breakage rate, the more ideal the harvesting operation.

[0125] The degree of crushing refers to the severity of the crushing of the kernels. It not only takes into account the number of crushing, but also the degree of damage to each crushed kernel. For example, if the kernel is broken into two halves, the degree of crushing is light; if it is crushed into powder, the degree of crushing is heavy. The degree of crushing reflects the specific situation of kernel breakage, which can help analyze the severity of crushing and then guide the adjustment of the harvester's working parameters to avoid excessive crushing.

[0126] Step 3: Mark the grains in the image data set with a second label according to the number and size of impurities contained, establish an impurity recognition model based on a convolutional neural network, and train it. Preprocess the real-time collected images, input them into the trained impurity recognition model, output the second label, and calculate the impurity rate.

[0127] By outputting the second label, the system first identifies and marks the impurities, and does not immediately calculate the impurity rate directly, but instead generates a label that can be used for subsequent processing. This method increases the flexibility of the system in subsequent processing because the output label can be adjusted, modified, or recalculated at multiple stages. For example, in subsequent steps, the impurity label can be corrected according to actual conditions to optimize the final impurity rate calculation. By outputting labels, multiple adjustments can be made according to changes in actual applications to improve the adaptability and accuracy of the system.

[0128] By marking the second label instead of directly calculating the impurity rate, the system can provide an intermediate result for the subsequent stage. This method can better verify, optimize and adjust the model during the training process. If there are recognition errors in the training data, users can improve the training set by adjusting the labels to prevent the system from being affected by erroneous data. In actual applications, real-time feedback labels can help optimize the operation of the harvester and further improve the quality of the work.

[0129] Convolutional neural networks are used to train the impurity recognition model. Through deep learning, multi-layer features in the image can be extracted, thereby improving the accuracy of impurity recognition. Traditional methods may rely on manual feature extraction and simple algorithms, which are easily affected by factors such as lighting and image quality. Convolutional neural networks can adaptively learn features, reduce human intervention, and improve the robustness of recognition. Traditional impurity recognition usually relies on threshold-based image processing methods, which are easily affected by background interference and require complex manual adjustments. Convolutional neural networks can automatically learn from large amounts of data, are highly adaptable, and can handle more complex impurity detection tasks.

[0130] In this embodiment, the training steps of the impurity recognition model are as follows: the image data set is used as the input of the impurity recognition model, and the second label is used as the output of the impurity recognition model to train the impurity recognition model.

[0131] The calculation formula of the impurity content is:

[0132]

[0133] Among them, Phz is the impurity rate, Sxs is the pixel area of ​​the image, Szz ia is the pixel area of ​​the iath impurity;

[0134] During harvesting operations, the harvested grains should be free of impurities as much as possible. Excessive impurities will affect the purity and quality of the grains, thus reducing their market value. Therefore, accurate assessment of the impurity content helps to judge the quality of the operation, especially whether the harvester effectively removes impurities. The higher the impurity content, the worse the quality of the harvester's operation.

[0135] By establishing two independent recognition models, the grain state recognition model and the impurity recognition model, first of all, the characteristics of grain breakage state and impurities are quite different, involving different visual information and calculation methods. The grain breakage state mainly focuses on the shape, degree of breakage, and damaged parts of the grain, while the impurities focus on the type, shape, and size of the impurities. Therefore, processing these two types of tasks separately can more effectively improve the recognition accuracy of each task.

[0136] Establishing models separately can ensure that each model focuses on processing specific tasks, so that the model of each task can perform deeper learning and optimization in its specific field. For example, the grain state recognition model can focus on refining the evaluation of the degree of breakage, while the impurity recognition model focuses on accurately detecting various impurities. This can avoid the interference between tasks and the decrease in recognition accuracy caused by excessive complexity when a model processes multiple tasks at the same time. Secondly, the two independent models can select more appropriate network structures, feature extraction methods and parameter settings according to the characteristics of their respective tasks, which helps to further improve the overall recognition accuracy and stability.

[0137] The use of two models for specialized processing can effectively reduce the risk of misidentification and provide higher robustness in complex environments. For example, the appearance of impurities and broken kernels is quite different. If a single model is used for mixed identification, the model will easily confuse the two types of objects, resulting in misclassification and inaccurate evaluation results. After establishing two models separately, the system can visually identify each type of object more clearly, reduce errors, and improve the accuracy and reliability of the evaluation. In addition, such a model structure is also more flexible, which is convenient for adjustment and optimization in the future according to different crops, environments or needs.

[0138] Step 4: Obtain historical environmental data and material properties of cereal straw, discretize the historical environmental data according to the time interval between photographing, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, take the grain breakage rate and degree of crushing, as well as the adjacent values ​​of the coordinate position at the time of image acquisition as features, classify the grain distribution, and calculate the uniformity of grain distribution based on the classification results.

[0139] In traditional harvester operation quality assessment, the focus is usually on factors such as equipment performance and crop characteristics, but the interference of environmental factors on operation results is ignored. By introducing historical environmental data and combining the material characteristics of grain straw, the impact of the environment can be quantified, thereby more comprehensively evaluating the harvester's operation quality. The environmental interference coefficient can more accurately judge the harvester's performance in a specific environment, thereby more accurately evaluating the harvesting quality.

[0140] The introduction of the environmental interference coefficient can help accurately quantify the impact of environmental factors such as humidity, temperature, wind speed, etc. on the quality of harvester operation. By calculating the environmental interference coefficient, it is possible to clearly understand the changes in the impurity content of the harvester under different environmental conditions.

[0141] For example, in an environment with high humidity, the moisture content of grains may be high, resulting in incomplete impurity separation and increased impurity content. The environmental interference factor can take these factors into account and determine how these external environmental conditions directly affect the separation of impurities during crop harvesting. In windy or temperature-difference environments, when the harvester has a strong anti-interference ability, it can effectively resist the impact of environmental changes on the quality of the operation and ensure that the removal of impurities is not significantly affected by external interference.

[0142] In this embodiment, the environmental data includes temperature value, humidity value and wind speed value;

[0143] The calculation formula for calculating the environmental interference coefficient is:

[0144]

[0145] Among them, Phg is the environmental interference coefficient, Wsd is the humidity value, Twd is the temperature value, ka is the correction coefficient, Vfs is the wind speed value, Pcz is the material characteristic of the straw, and t is the current time interval.

[0146] By introducing multiple features such as grain breakage rate, degree of crushing, and adjacent values ​​of coordinate positions for comprehensive evaluation, a more comprehensive perspective on operation quality analysis is provided. This multi-dimensional analysis can more accurately reflect the actual performance of the harvester during operation. By classifying the distribution of grains and further calculating the uniformity of distribution, the operation quality of the harvester can be more accurately evaluated, and it can be determined which areas have good operation effects and which areas may have problems during the harvesting process. The gap between the initial distribution of grains and the crushing rate and degree of crushing after harvesting is a key factor in measuring the quality of operation. This step can more clearly understand whether the harvester can harvest evenly in different areas by calculating the uniformity of grain distribution, and whether there are some areas where grain losses are serious due to improper operation of the harvester. In actual operations, the operation quality of the harvester is often affected by regional factors, such as terrain changes, crop density, and improper adjustment of the harvester. Analyzing the uniformity of distribution can help judge the quality of the operation.

[0147] The calculation formula of the adjacent value of the coordinate position is:

[0148]

[0149] Among them, PzbDlj is the adjacent value of the grain coordinate position, (x iq ,y iq ), (x ip ,y ip ) are the coordinate positions of the iqth and ipth seeds respectively.

[0150] The breakage rate and degree of crushing are direct indicators for evaluating the quality of operation and can reflect the effect of the harvester during the harvesting process. The proximity of the coordinate position serves as a spatial feature to help identify the operating effects of different areas and provide information in the spatial dimension. For example, some areas may have serious grain damage due to uneven machine operation. The proximity of the coordinate position can be used to analyze the specific problems in these areas. The combination of these three can more comprehensively reflect the quality of the harvester's operation, because they conduct a multi-angle quantitative evaluation of the operating effect from the two dimensions of physical state and spatial position.

[0151] The clustering method classifies the characteristics of grain distribution, so that when evaluating the quality of the operation, the distribution of grains can be analyzed more carefully. It can not only analyze the overall breakage rate, but also gain in-depth understanding of the detailed differences in the operation process by identifying the distribution of different categories. The clustering results help reveal which areas have uneven grain distribution and severe breakage, as well as the differences between these areas and the overall distribution. This analysis method is more accurate than relying solely on averages or overall indicators. The clustering method can identify differences in operations in different areas and times, avoiding errors that may occur in traditional methods (such as ignoring differences in local areas). This combination of space and degree of breakage can more accurately evaluate the quality of the entire operation process.

[0152] In this embodiment, the specific steps of classifying the grain distribution are:

[0153] Select the fragmentation rate Lps, fragmentation degree Pps, and proximity value Pzb as the feature vector of the coordinate position corresponding to each image recognition moment;

[0154] Randomly select K data nodes as the initial centroids, where K represents the number of categories after clustering and K is a positive integer. The i-th initial centroid is represented as: C i [Lps(i), Pps(i), Pzb(i)], Lps(i), Pps(i) and Pzb(i) respectively represent the fragmentation rate, fragmentation degree and proximity value of the i-th data node as the initial centroid, i is a positive integer, and i = 1, 2, ..., K;

[0155] For each feature vector at each coordinate position, calculate its distance to each initial centroid and assign it to the cluster represented by the nearest centroid. The formula for calculating the distance to the initial centroid is:

[0156]

[0157] Where d(i, j) represents the distance between the i-th initial centroid and the j-th data node, Lps(j), Pps(j) and Pzb(j) represent the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the j-th data node, respectively;

[0158] For each cluster, after each clustering is completed, the mean of all points in the cluster is recalculated, and this mean is used as the feature data of the new centroid. The update formula of the i-th centroid feature data is:

[0159]

[0160] Where m represents the number of data nodes assigned to the i-th centroid, Lps(i) old 、Pps(i) oldand Pzb(i) old Indicates the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the data node assigned to the i-th centroid, Lps(i) new 、Pps(i) new and Pzb(i) new Indicates the fragmentation rate, fragmentation degree and proximity value of the updated centroid;

[0161] Based on the updated centroid feature vector, clustering is performed again until the change in the position of all centroids is less than the threshold, then the clustering is considered stable and the clustering is terminated;

[0162] The calculation formula for calculating the uniformity of grain distribution according to the classification results is:

[0163]

[0164] Among them, Pdx is the uniformity of grain distribution, nz ik is the number of coordinate points in the ikth cluster, and K is the number of clusters.

[0165] Grain distribution uniformity refers to whether the spatial distribution of grain harvesting status in the operating area is uniform during the harvesting process. If the grain distribution is very uniform, the harvesting effect of the harvester in each area during the operation should be relatively consistent, and the breakage rate and degree of crushing will be relatively balanced. If the grain distribution is uneven, the harvesting effect of the harvester in each area during the operation should vary greatly, and the breakage rate and degree of crushing will also fluctuate greatly.

[0166] Kernel distribution uniformity helps understand the breakage or crushing of kernels during harvester operation. For example, the breakage rate may be higher in areas with uneven distribution, while the breakage rate may be lower in areas with uniform distribution. By combining distribution uniformity with breakage and crushing data, the overall performance of the harvester can be more accurately assessed.

[0167] Step 5: Calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals. Calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals. Judge the operation quality based on the anti-interference coefficient and grain harvesting quality.

[0168] Indicators such as environmental interference coefficient and impurity content can reflect the impact of external factors on the operation process. This evaluation method that takes environmental interference into account ensures that the harvester's operation quality is not only limited to the performance of the machine itself, but also takes into account the interference of external factors.

[0169] Existing technologies may focus more on the performance and working status of the machine itself, while ignoring the impact of changes in the environment and working conditions on the quality of work. This patented solution introduces an anti-interference coefficient, which enables the evaluation system to flexibly respond to external changes, thereby more accurately judging the quality of work.

[0170] The anti-interference coefficient can quantify the adaptability and stability of the harvester under different environmental conditions. For example, whether the harvester can maintain stable operation quality in a bad weather environment. This feature is very important because in reality, environmental factors often have a significant impact on the operation results. By calculating the anti-interference coefficient, the performance of the harvester in a complex environment can be evaluated, which helps optimize the operation strategy.

[0171] In this embodiment, the calculation formula for calculating the anti-interference coefficient of the harvester is:

[0172]

[0173] Wherein, Gkg is the anti-interference coefficient, Phg is the environmental interference coefficient, Phz is the impurity rate, ic is the icth captured image, and nc is the number of image captures.

[0174] The anti-interference coefficient is an indicator used to measure the adaptability and stability of the harvester under different environmental interference conditions. The higher the anti-interference coefficient, the stronger the harvester's ability to adapt to environmental changes and the less susceptible to external interference during operation. On the contrary, a low anti-interference coefficient may mean that under complex environmental conditions, the harvester may not be able to maintain a stable working state and the operating quality may be reduced.

[0175] In this embodiment, the calculation formula for calculating the grain harvesting quality is:

[0176]

[0177] Among them, Lpsh is the stable value of the breakage rate, Ppsh is the stable value of the breakage degree, Lps is the breakage rate, Pps is the breakage degree, ic is the icth image collected, nc is the number of image collections, Pdx is the uniformity of grain distribution, and Gzz is the grain harvesting quality.

[0178] Grain harvesting quality refers to the harvesting effect of the harvester on the grains during the operation, including the degree of damage, breakage rate and distribution uniformity of the grains. It comprehensively evaluates the operation quality by calculating the breakage rate, degree of damage and distribution uniformity of the grains.

[0179] The quality of grain harvesting directly affects the efficiency of harvesting and the final yield. If the breakage rate and degree of crushing are low, it means that the harvester is operated properly, the crop loss is small, and the harvesting quality is good; if the grain distribution is uneven, or the crushing and breakage are serious, it may indicate that there are problems in the harvester operation and adjustments and optimizations are needed.

[0180] By calculating the breakage rate and damage degree at different time intervals, a dynamic feedback on the operation quality can be provided. This means that during the operation, the system can adjust the quality assessment in real time according to the time changes, further improving the accuracy and timeliness of the assessment.

[0181] Traditional technologies may only rely on one-time or static evaluations, and cannot provide real-time feedback and dynamic optimization. However, this solution can provide more detailed operation quality monitoring based on environmental changes and harvesting status in different time periods.

[0182] By combining the uniformity of grain distribution with the environmental interference coefficient, not only can the changes in harvesting quality be accurately calculated, but also the sudden interference factors during the operation can be responded to in a timely manner. For example, in the case of large environmental interference, the distribution of grains may become uneven, resulting in increased breakage or crushing rates. Taking these factors into consideration, the operation quality of the harvester can be evaluated more accurately.

[0183] In this embodiment, the calculation formula for judging the operation quality based on the anti-interference coefficient and the grain harvesting quality is:

[0184] Gl=Gkg*α1+Gzz*α2

[0185] Among them, Gl is the operation quality of the harvester, α1 and α2 are the weights of the anti-interference coefficient and grain harvesting quality, α1+α2=1, 0<α1<1, 0<α2<1

[0186] See also Figure 2 The present invention also provides a harvester operation quality assessment system based on grain status, wherein the harvester operation quality assessment system is used to execute the above-mentioned harvester operation quality assessment method based on grain status, comprising:

[0187] A data acquisition module, which is used to obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval when the harvester is operating, and preprocess the images to form an image data set, and establish a coordinate system by recording the harvesting trajectory when harvesting grains, and making a one-to-one correspondence between the position of the harvester at the time of image acquisition and the acquired image;

[0188] A grain recognition module, wherein the grain recognition module is used to mark the grains in the image data set with a first label according to the grain breakage state and severity, establish a grain state recognition model based on a convolutional neural network, and train the model, pre-process the real-time collected image, input it into the trained grain state recognition model, output the first label, and calculate the breakage rate and breakage degree;

[0189] An impurity identification module is used to mark the kernels in the image data set with a second label according to the number and size of impurities contained, establish an impurity identification model based on a convolutional neural network, and perform training, pre-process the real-time collected image, input it into the trained impurity identification model, output the second label, and calculate the impurity content;

[0190] An interference analysis module, which is used to obtain historical environmental data and material properties of grain straw, discretize the historical environmental data according to the time interval between photographing, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, classify the grain distribution by taking the grain breakage rate and degree of breakage, and the adjacent values ​​of the coordinate position at the time of image acquisition as features, and calculate the uniformity of grain distribution according to the classification results;

[0191] The quality assessment module is used to calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals, calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals, and judge the operation quality according to the anti-interference coefficient and the grain harvesting quality.

[0192] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0193] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0194] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0195] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for evaluating harvester operation quality based on grain status, characterized in that: The specific steps include: Step 1: Obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval, preprocess the images to form an image data set, and establish a coordinate system by recording the harvesting trajectory when harvesting grains and making a one-to-one correspondence between the harvester's position at the time of image acquisition and the acquired image; Step 2: The kernels in the image data set are labeled with a first label according to the kernel breakage state and severity, a kernel state recognition model is established based on a convolutional neural network, and the model is trained. After preprocessing the real-time collected images, the images are input into the trained kernel state recognition model, the first label is output, and the breakage rate and degree of breakage are calculated; Step 3: Mark the kernels in the image data set with a second label according to the number and size of impurities contained, establish an impurity recognition model based on a convolutional neural network, and train it. Preprocess the real-time collected images, input them into the trained impurity recognition model, output the second label, and calculate the impurity rate. Step 4: Obtain historical environmental data and material properties of grain straw, discretize the historical environmental data according to the time interval between photos, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, take the grain breakage rate and degree of breakage, and the adjacent values ​​of the coordinate position at the time of image acquisition as features, classify the grain distribution, and calculate the uniformity of grain distribution based on the classification results; Step 5: Calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals. Calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals. Judge the operation quality based on the anti-interference coefficient and grain harvesting quality.

2. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The method for preprocessing the image is image normalization, and the specific calculation formula is: The method for preprocessing the image is to normalize the pixels of the image. The calculation formula is: Among them, Xsg is the normalized pixel value, and Xs is the pixel value of the image.

3. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The convolutional neural network specifically includes a convolution layer, an activation layer, a pooling layer and an output layer; Convolutional layer: extracts features from the bounding box and label of the input image through convolution operation, and the calculation formula is: Where I is the input image matrix, K is the convolution kernel, (xc, yc) is the coordinate of the output matrix, and K(ic, jc) is the value of the convolution kernel in the icth row and jcth column; The activation layer is calculated as: ReLU(xz)=max(0,xz) Among them, xz is the coordinate of the output matrix of the convolutional layer output; The calculation formula of the pooling layer is: MaxPooling(xd,yd)=max{I(xc+ic,yc+jc)|ic,jc∈(3,3)} Where I is the input matrix, MaxPooling(xd,yd) is the maximum pooling output; The calculation formula of the output layer is: y=W*xm+b Among them, xm is the output of the pooling layer, y is the result of image recognition, W is the weight, and b is the bias parameter; The image dataset is used as the input of the grain state recognition model, the first label is used as the output of the grain state recognition model, and the grain state recognition model is trained.

4. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The seed crushing state is divided into epidermal crushing and whole crushing; The severity levels are classified into slight epidermal fragmentation, severe epidermal fragmentation, slight overall fragmentation, and complete overall fragmentation; The label content includes the state, severity and number of kernels broken; The calculation formula for calculating the crushing rate is: Among them, Nbp is the number of kernels with broken skin, Nzp is the number of kernels with broken whole, Nzl is the number of all kernels, and Lps is the breakage rate; The calculation formula for calculating the degree of fragmentation is: Among them, Pps is the degree of breakage, Nbp1 and Nbp2 are the number of kernels with slightly broken skin and severely broken skin, respectively, Nzp1 and Nzp2 are the number of kernels with slightly broken skin and completely broken skin, respectively, α1 is the influence coefficient of kernels with broken skin, and α2 is the influence coefficient of kernels with completely broken skin.

5. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The training steps of the impurity recognition model are as follows: the image data set is used as the input of the impurity recognition model, and the second label is used as the output of the impurity recognition model to train the impurity recognition model; The calculation formula of the impurity content is: Among them, Phz is the impurity rate, Sxs is the pixel area of ​​the image, Szz ia is the pixel area of ​​the iath impurity.

6. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The environmental data includes temperature value, humidity value and wind speed value; The calculation formula for calculating the environmental interference coefficient is: Among them, Phg is the environmental interference coefficient, Wsd is the humidity value, Twd is the temperature value, ka is the correction coefficient, Vfs is the wind speed value, Pcz is the material characteristic of the straw, and t is the current time interval; The calculation formula of the adjacent value of the coordinate position is: Among them, PzbDlj is the adjacent value of the grain coordinate position, (x iq ,y iq ), (x ip ,y ip ) are the coordinate positions of the iqth and ipth seeds respectively.

7. The method for evaluating harvester operation quality based on grain status according to claim 6, characterized in that: The specific steps of classifying the grain distribution are: Select the fragmentation rate Lps, fragmentation degree Pps, and proximity value Pzb as the feature vector of the coordinate position corresponding to each image recognition moment; Randomly select K data nodes as the initial centroids, where K represents the number of categories after clustering and K is a positive integer. The i-th initial centroid is represented as: C i [Lps(i), Pps(i), Pzb(i)], Lps(i), Pps(i) and Pzb(i) respectively represent the fragmentation rate, fragmentation degree and proximity value of the i-th data node as the initial centroid, i is a positive integer, and i = 1, 2, ..., K; For each feature vector at each coordinate position, calculate its distance to each initial centroid and assign it to the cluster represented by the nearest centroid. The formula for calculating the distance to the initial centroid is: Where d(i, j) represents the distance between the i-th initial centroid and the j-th data node, Lps(j), Pps(j) and Pzb(j) represent the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the j-th data node, respectively; For each cluster, after each clustering is completed, the mean of all points in the cluster is recalculated, and this mean is used as the feature data of the new centroid. The update formula of the i-th centroid feature data is: Where m represents the number of data nodes assigned to the i-th centroid, Lps(i) old 、Pps(i) old and Pzb(i) old Indicates the fragmentation rate, fragmentation degree and proximity value of the coordinate position corresponding to the data node assigned to the i-th centroid, Lps(i) new 、Pps(i) new and Pzb(i) new Indicates the fragmentation rate, fragmentation degree and proximity value of the updated centroid; Based on the updated centroid feature vector, clustering is performed again until the change in the position of all centroids is less than the threshold, then the clustering is considered stable and the clustering is terminated; The calculation formula for calculating the uniformity of grain distribution according to the classification results is: Among them, Pdx is the uniformity of grain distribution, nz ik is the number of coordinate points in the ikth cluster, and K is the number of clusters.

8. The method for evaluating harvester operation quality based on grain status according to claim 1, characterized in that: The calculation formula for calculating the anti-interference coefficient of the harvester is: Among them, Gkg is the anti-interference coefficient, Phg is the environmental interference coefficient, Phz is the impurity rate, ic is the image collected at the icth time, and nc is the number of image collections; The calculation formula for calculating the grain harvest quality is: Among them, Lpsh is the stable value of the breakage rate, Ppsh is the stable value of the breakage degree, Lps is the breakage rate, Pps is the breakage degree, ic is the image collected at the icth time, nc is the number of image collections, Pdx is the uniformity of grain distribution, and Gzz is the grain harvesting quality; the calculation formula for judging the operation quality based on the anti-interference coefficient and the grain harvesting quality is: Gl=Gkg*α1+Gzz*α2 Among them, Gl is the operating quality of the harvester, α1 and α2 are the weights of the anti-interference coefficient and the grain harvesting quality, respectively, and α1+α2=1.

9. A harvester operation quality assessment system based on grain status, characterized in that: The harvester operation quality assessment system is used to execute the harvester operation quality assessment method based on grain status according to any one of claims 1 to 8, comprising: A data acquisition module, which is used to obtain historical grain images in the storage bin of the harvester to be evaluated at the same time interval when the harvester is operating, and preprocess the images to form an image data set, and establish a coordinate system by recording the harvesting trajectory when harvesting grains, and making a one-to-one correspondence between the position of the harvester at the time of image acquisition and the acquired image; A grain recognition module, wherein the grain recognition module is used to mark the grains in the image data set with a first label according to the grain breakage state and severity, establish a grain state recognition model based on a convolutional neural network, and train the model, pre-process the real-time collected image, input it into the trained grain state recognition model, output the first label, and calculate the breakage rate and breakage degree; An impurity identification module is used to mark the kernels in the image data set with a second label according to the number and size of impurities contained, establish an impurity identification model based on a convolutional neural network, and perform training, pre-process the real-time collected image, input it into the trained impurity identification model, output the second label, and calculate the impurity content; An interference analysis module, which is used to obtain historical environmental data and material properties of grain straw, discretize the historical environmental data according to the time interval between photographing, calculate the environmental interference coefficient through the discretized historical environmental data and material properties, classify the grain distribution by taking the grain breakage rate and degree of breakage, and the adjacent values ​​of the coordinate position at the time of image acquisition as features, and calculate the uniformity of grain distribution according to the classification results; The quality assessment module is used to calculate the anti-interference coefficient of the harvester through the environmental interference coefficient and impurity content at different time intervals, calculate the grain harvesting quality according to the uniformity of grain distribution and the breakage rate and damage degree at different time intervals, and judge the operation quality according to the anti-interference coefficient and the grain harvesting quality.

Images