Method and system for identifying maturity of leaf crops based on small sample images

Abstract

The invention discloses a method and a system for identifying the maturity of leaf crops based on small sample images, mainly relates to the technical field of identifying and harvesting the maturity, and is used for solving the problems that an existing identification scheme needs more training samples and needs to consume more manpower and material resources. Comprising the following steps: extracting Red, Green and Blue channels of a third preset number of leaves of each preset maturity in a sample group and a color gradation mean value of a grayscale image, and forming a table corresponding to each preset maturity; on the basis of the color gradation mean value, a pixel union set method is adopted to carry out union set operation on the tables corresponding to the preset maturity degrees, and union set data sets corresponding to the Red, Green and Blue channels and gray level images in the preset maturity degrees are obtained; extracting a feature interval of the union set data set; and taking the endpoint value of the feature interval as training data to obtain a trained Bayesian model.

Description

Technical Field

[0001] The present application relates to the technical field of identifying the maturity of agricultural products, and in particular to a method and system for identifying the maturity of leaf crops based on small sample images. Background Art

[0002] Harvest maturity refers to the maturity of agricultural products after they have completed their growth and material accumulation and have reached the maturity level that can be harvested. It is an important basis for determining the timing of agricultural product harvesting. Harvest maturity is closely related to the yield, quality, shelf life and profit of agricultural products. In field production, the identification of harvest maturity mainly depends on the experience of the harvesters, which is highly subjective and is greatly affected by factors such as variety, part, and light (Chen et al., 2019). How to objectively and accurately judge the harvest maturity of agricultural products has become the focus of attention of agricultural product purchasers and processors. At present, the methods for identifying the harvest maturity of agricultural products are mainly divided into chemical detection (Assefa et al., 2019), spectral analysis (Lee et al., 2020), and image recognition (Yuan et al., 2024). Due to the long processing time of chemical detection and the expensive and inconvenient collection of spectral analysis equipment, the application scope of both is greatly limited.

[0003] The maturity process of agricultural products is a process of change from the inside to the outside. The maturity state is not only manifested in the change of appearance shape and size, but also in the appearance and (or) change of characteristic color and texture. Among them, color information is an important characteristic parameter for the identification of agricultural product maturity, which can be directly collected through digital images. With the development of digital imaging, artificial intelligence and big data technology, the comprehensive use of digital images and machine learning technology has become the mainstream means of intelligent identification of agricultural product maturity. And with the introduction of the concept of big data, the number of input samples is also increasing. It is believed that the more samples collected, the closer the data set formed is to the real group, and the better the accuracy of the model constructed by this. However, this brings new research problems. The more samples collected, the more manpower and material resources are required, which leads to a rapid increase in the construction cost of the agricultural product harvest maturity model, which in turn restricts the further promotion and application of the recognition model. How to achieve the accuracy of the Bayesian model of agricultural product harvest maturity under small samples has become a very challenging and application-worthy topic. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present application provides a method and system for recognizing the maturity of leaf crops based on small sample images, so as to solve the problem that the existing recognition schemes require a large number of training samples and consume a lot of manpower and material resources.

[0005] In a first aspect, the present application provides a method for identifying the maturity of leaf crops based on a small sample image, the method comprising: Collect a second preset number of leaves from the bottom to the top of a first preset number of crop plants; divide the leaves into a sample group and a test group; wherein both the sample group and the test group are marked with preset maturity levels; extract the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level means, use the pixel union method to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity; extract the feature interval of the union data set; use the endpoint values ​​of the feature interval as training data to train a Bayesian model, and verify the trained Bayesian model based on the test group.

[0006] In one implementation of the present application, based on the color level mean, a pixel union method is used to perform a union operation on the tables corresponding to each preset maturity, and a union data set corresponding to each preset maturity of the Red, Green, Blue channels and grayscale images is obtained, specifically including: Using the [ ] operator of MATLAB software, the color level means of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity are respectively unioned to obtain the corresponding union data sets of the Red, Green, Blue channels and grayscale images at each preset maturity.

[0007] In one implementation of the present application, extracting the feature interval of the union data set specifically includes: The prctile function of MATLAB software was used to extract the percentiles Q1 and Q3 of the union data set corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity, and based on this, the surface color feature intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity were constructed.

[0008] In one implementation of the present application, extracting the feature interval of the union data set specifically includes: Using the quartile method, we obtain the values ​​at the 25% and 75% quantiles in the union data set, which are recorded as Q1 and Q3 respectively; Extract the interval [Q1,Q3] of the union dataset as the feature interval of the union dataset.

[0009] In one implementation of the present application, the feature interval endpoint values ​​are used as training data to train the Bayesian model, specifically including: The characteristic intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity are obtained, and the endpoint values ​​of the characteristic intervals are used as input data, and then the Bayesian judgment model is constructed using the Bayesian toolbox.

[0010] In one implementation of the present application, the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group are extracted to form a table, specifically including: Extracting the color level mean of the Red, Green, Blue channels and grayscale image of the third preset number of leaves of each preset maturity in the sample group; Extracting the color level mean of the Red, Green, Blue channels and grayscale images of the fourth preset number of leaves of each preset maturity in the sample group; wherein the third preset number is less than the fourth preset number; Calculating a value difference rate between the color level means corresponding to the third preset number and the fourth preset number; When the value difference rate is greater than a preset threshold, the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group are re-extracted until the value difference rate between the color level means corresponding to the third preset number and the fourth preset number is less than or equal to the preset threshold; Generates a table of the mean values ​​of the red, green, blue channels and the grayscale image whose value difference ratio is less than or equal to the preset threshold.

[0011] In one implementation of the present application, the trained Bayesian model is verified based on the test group, specifically including: The accuracy of the Bayesian model is tested through the test group. When the accuracy is less than the preset accuracy threshold, the third preset number is updated based on the preset growth rate, and the Bayesian model is obtained again until the accuracy is greater than or equal to the preset accuracy threshold, thereby obtaining a trained Bayesian model.

[0012] In a second aspect, the present application provides a system for recognizing the maturity of leaf crops based on a small sample image, the system comprising: A collection module, used for collecting a second preset number of leaves from the first preset number of crop plants from bottom to top; dividing the leaves into a sample group and a test group; wherein both the sample group and the test group are marked with a preset maturity degree; The union acquisition module is used to extract the color level mean of the Red, Green, Blue channels and grayscale images of the third preset number of leaves at each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level mean, the pixel union method is used to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity; The model building module is used to extract the feature interval of the union data set; the feature interval endpoint values ​​are used as training data to train the Bayesian model, and the trained Bayesian model is verified based on the test group.

[0013] In one implementation of the present application, the union acquisition module includes a union unit, The [ ] operator of MATLAB software is used to perform union operations on the color level means of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity, and obtain the corresponding union data sets of the Red, Green, Blue channels and grayscale images at each preset maturity.

[0014] Those skilled in the art can understand that the present application has at least the following beneficial effects: The present application provides a method and system for identifying the maturity of leaf crops based on small sample images. Traditional methods usually require a large number of samples for training, which leads to high manpower and material costs. The present application significantly reduces the number of samples required by only collecting a first preset number of crop plant leaves and dividing them into a sample group and a test group, thereby reducing the collection cost. In addition, the present application involves: extracting the RGB channel of the leaf and the color level mean of the grayscale image, forming a table corresponding to the preset maturity, and merging the table by the pixel union method to obtain the feature interval. This process simplifies the data preprocessing steps and improves the efficiency of model construction. In addition, by training the Bayesian model based on the feature interval using the interval judgment method, the present application can better capture the key differences between leaves of different maturity, thereby enhancing the generalization ability of the model. This enables the model to achieve a high recognition accuracy even with a small sample.

[0015] That is, this application uses RGB digital image pixels as analysis units and adopts union operations to reconstruct the group information data set to maximize the acquisition of true and accurate surface color information of agricultural product groups while reducing the number of sampling (cost), thereby constructing a more accurate agricultural product maturity determination model.

[0016] In addition, in order to avoid the impact of too few samples on the accuracy of the model, this application introduces a dynamic optimization mechanism, that is, when the model accuracy tested by the test group is less than the preset accuracy threshold, the first preset number will be updated according to the preset growth rate, and the model will be retrained. This mechanism ensures that the model can gradually improve its accuracy in continuous iteration until it meets the application requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which: Figure 1 This is a flow chart of a method for identifying the maturity of leaf crops based on small sample images provided in an embodiment of the present application.

[0018] Figure 2This is a schematic diagram of the internal structure of a system for recognizing the maturity of leaf crops based on small sample images provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through the preferred embodiments. The preferred embodiments are only used to explain the technical principles of the present disclosure, and are not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should still fall within the protection scope of the present disclosure.

[0020] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0021] How to achieve the accuracy of the Bayesian model of agricultural product harvest maturity under small samples has become a very challenging and valuable topic. To solve this problem, the first thing is to require researchers to extract as much group characteristic information as possible from small samples. The imaging unit of digital images is pixels, which makes the physically separated agricultural products have underlying associations. Therefore, we consider whether we can try to reconstruct the group information data set from the pixel unit scale of the digital image of agricultural products, so as to obtain the true and accurate surface color information of the agricultural product group to the greatest extent under the condition of reducing the number of samples (cost), so as to build a high-precision agricultural product maturity determination model.

[0022] Tobacco is a model plant widely used in genetic and ecological research. Due to its large leaf area and overall flatness, the surface color of the leaves changes from green to yellow during maturity. It is an ideal material for obtaining surface color information and maturity analysis, and can provide valuable reference for the construction of Bayesian models for harvest maturity of other leaf agricultural products. The Bayesian classifier is a probabilistic classifier based on the naive Bayes theorem. It is used to classify the color features of digital images of agricultural products, and its performance is better than that of multi-layer perceptron, fuzzy logic, and principal component analysis methods. Therefore, this application takes fresh tobacco leaves as the research object, obtains the digital information of the surface color of fresh tobacco leaves with three degrees of maturity: under-ripe, mature, and over-ripe, adopts the pixel union method to obtain the color information data sets of fresh tobacco leaves with different degrees of maturity, and analyzes the statistical characteristics of these data sets; then, the data fitting method, the single leaf set interval judgment method, and the pixel union interval judgment method are used to construct tobacco maturity classification models with different sample sizes. By comparing the model judgment accuracy, the advantages of the pixel union interval judgment method are clarified, the construction mechanism of the precise model of small sample agricultural product harvest maturity is explained, and a research case is provided.

[0023] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0024] The embodiment provides a method for identifying the maturity of leaf crops based on a small sample image. Figure 1 As shown, the method provided in the embodiment of the present application mainly includes the following steps: Step 110: Collect a second preset number of leaves from the first preset number of crop plants from bottom to top; and divide the leaves into a sample group and a test group.

[0025] It should be noted that the sample group and the test group are both marked with preset maturity levels. The number of leaves at each preset maturity level in the sample group is the same.

[0026] In some embodiments, this step can be to select 50 tobacco plants with normal growth and consistent development, collect leaves at the 6th to 11th leaf positions from the bottom to the top (6 leaves per plant), remove damaged and incomplete single leaves, totaling 272 leaves, and randomly divide tobacco leaves of different maturity into two groups, the first group is used as the modeling data source, including 50 samples each of unripe, mature and overripe (a total of 150 samples), and the second group is used as the test data source, including 28 unripe samples, 50 mature samples and 44 overripe samples (a total of 122 samples).

[0027] Step 120, extract the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves at each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level means, use the pixel union method to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity.

[0028] It should be noted that this step can be specifically to use the mean function of MATLAB software to extract the color level mean of the Red, Green, Blue channels and grayscale images of all collected leaves. Randomly extract the color level mean of the Red, Green, Blue channels and grayscale images of each third preset number (for example, 5 single leaves) of each maturity type from the sample group to form a table corresponding to each preset maturity. Then, using the "[ ]" operator of MATLAB software, randomly extract the second preset number of leaves of each maturity type from the sample group for union operation, obtain the group pixel union data set of the three color channels and grayscale images under the three maturity levels, and form a union data set.

[0029] In addition, in order to reduce the mixing of fault data during the extraction process, the present application can extract the color level mean of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group, and form a table to verify the data. The specific process can be: Extracting the color level mean of the Red, Green, Blue channels and grayscale image of the third preset number of leaves of each preset maturity in the sample group; Extracting the color level mean of the Red, Green, Blue channels and grayscale images of the fourth preset number of leaves of each preset maturity in the sample group; wherein the third preset number is less than the fourth preset number; Calculating a value difference rate between the color level means corresponding to the third preset number and the fourth preset number; When the value difference rate is greater than a preset threshold, the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group are re-extracted until the value difference rate between the color level means corresponding to the third preset number and the fourth preset number is less than or equal to the preset threshold; Generates a table of the mean values ​​of the red, green, blue channels and the grayscale image whose value difference ratio is less than or equal to the preset threshold.

[0030] Those skilled in the art can understand that, no matter the number of samples is 5, 10 or 30, the difference rate between the mean of the Red channel, Green channel, Blue channel, and Gray image obtained after the pixel union and the corresponding eigenvalue after the sample union is within 5%, and the difference is small. Therefore, the characteristic of small difference can be used to verify whether the current data is mixed with abnormal data.

[0031] Step 130, extracting the characteristic interval of the union data set; using the characteristic interval endpoint values ​​as training data, training the Bayesian model, and verifying the trained Bayesian model based on the test group.

[0032] It should be noted that the feature interval of the extracted union data set can be specifically: The prctile function of MATLAB software was used to extract the percentiles Q1 and Q3 of the union data set corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity, and based on this, the surface color feature intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity were constructed.

[0033] In addition, Q1 and Q3 can be determined by those skilled in the art according to actual conditions.

[0034] As an example, the determination scheme of Q1 and Q3 may be: Using the quartile method, the 25% and 75% quantile values ​​of the union data set are obtained, which are recorded as Q1 and Q3 respectively; the interval [Q1, Q3] of the union data set is extracted as the characteristic interval of the union data set.

[0035] Among them, the feature interval endpoint values ​​are used as training data to train the Bayesian model, which can be specifically: The characteristic intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity are obtained, and the endpoint values ​​of the characteristic intervals are used as input data, and then the Bayesian judgment model is constructed using the Bayesian toolbox.

[0036] Among them, according to the test group, the trained Bayesian model is verified, including: The accuracy of the Bayesian model is tested through the test group. When the accuracy is less than the preset accuracy threshold, the third preset number is updated based on the preset growth rate, and the Bayesian model is obtained again until the accuracy is greater than or equal to the preset accuracy threshold, thereby obtaining a trained Bayesian model.

[0037] In addition, the present application can update the Bayesian model according to the accuracy rate. Specifically, the accuracy of the Bayesian model is tested by the test group, and when the accuracy is less than the preset accuracy threshold, the third preset number is updated based on the preset growth rate, and the trained Bayesian model is obtained again until the accuracy is greater than or equal to the preset accuracy threshold.

[0038] It should be noted that the preset growth rate can be 10.

[0039] Those skilled in the art can understand that in order to avoid the accuracy of the model being affected by too few samples, the present application introduces a dynamic optimization mechanism, that is, when the accuracy of the model tested by the test group is less than the preset accuracy threshold, the first preset number will be updated according to the preset growth rate, and the model will be retrained. This mechanism ensures that the accuracy of the model can be gradually improved in continuous iteration until it meets the application requirements.

[0040] In addition, this application Figure 2 The present application provides a system for recognizing the maturity of leaf crops based on small sample images. Figure 2 As shown, the system provided in the embodiment of the present application mainly includes: The collection module 210 is used to collect a second preset number of leaves from the first preset number of crop plants from bottom to top; divide the leaves into a sample group and a test group; wherein both the sample group and the test group are marked with a preset maturity degree.

[0041] The union acquisition module 220 is used to extract the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves at each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level means, the pixel union method is used to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity.

[0042] The union acquisition module 220 includes a union unit, The [ ] operator of MATLAB software is used to perform union operations on the color level means of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity, and obtain the corresponding union data sets of the Red, Green, Blue channels and grayscale images at each preset maturity.

[0043] The model building module 230 is used to extract the characteristic interval of the union data set; use the characteristic interval endpoint values ​​as training data to train the Bayesian model, and verify the trained Bayesian model based on the test group.

[0044] So far, the technical solutions of the present disclosure have been described in combination with the above multiple embodiments, but it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.

Claims

1. A method for identifying the maturity of leaf crops based on small sample images, characterized in that: The method comprises: Collecting a second preset number of leaves from the first preset number of crop plants from bottom to top; dividing the leaves into a sample group and a test group; wherein both the sample group and the test group are marked with a preset maturity degree; Extract the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves at each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level means, use the pixel union method to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity; Extract the feature interval of the union data set; use the endpoint values ​​of the feature interval as training data to train the Bayesian model, and verify the trained Bayesian model based on the test group.

2. The method for recognizing the maturity of leaf crops based on small sample images according to claim 1, characterized in that: Based on the color level mean, the pixel union method is used to perform a union operation on the tables corresponding to each preset maturity, and the union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity is obtained, including: Using the [ ] operator of MATLAB software, the color level means of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity are respectively unioned to obtain the corresponding union data sets of the Red, Green, Blue channels and grayscale images at each preset maturity.

3. The method for recognizing the maturity of leaf crops based on small sample images according to claim 1, characterized in that: Extract the feature intervals of the union data set, including: The prctile function of MATLAB software was used to extract the percentiles Q1 and Q3 of the union data set corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity, and based on this, the surface color feature intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity were constructed.

4. The method for recognizing the maturity of leaf crops based on small sample images according to claim 3, characterized in that: Extract the feature intervals of the union data set, including: Using the quartile method, we obtain the values ​​at the 25% and 75% quantiles in the union data set, which are recorded as Q1 and Q3 respectively; Extract the interval [Q1,Q3] of the union dataset as the feature interval of the union dataset.

5. The method for recognizing the maturity of leaf crops based on small sample images according to claim 1, characterized in that: The feature interval endpoint values ​​are used as training data to train the Bayesian model, including: The characteristic intervals corresponding to the Red, Green, Blue channels and grayscale images at each preset maturity are obtained, and the endpoint values ​​of the characteristic intervals are used as input data, and then the Bayesian judgment model is constructed using the Bayesian toolbox.

6. The method for recognizing the maturity of leaf crops based on small sample images according to claim 1, characterized in that: Extract the color level mean of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group to form a table, specifically including: Extracting the color level mean of the Red, Green, Blue channels and grayscale image of the third preset number of leaves of each preset maturity in the sample group; Extracting the color level mean of the Red, Green, Blue channels and grayscale images of the fourth preset number of leaves of each preset maturity in the sample group; wherein the third preset number is less than the fourth preset number; Calculating a value difference rate between the color level means corresponding to the third preset number and the fourth preset number; When the value difference rate is greater than a preset threshold, the color level means of the Red, Green, Blue channels and grayscale images of the third preset number of leaves of each preset maturity in the sample group are re-extracted until the value difference rate between the color level means corresponding to the third preset number and the fourth preset number is less than or equal to the preset threshold; Generates a table of the mean values ​​of the red, green, blue channels and the grayscale image whose value difference ratio is less than or equal to the preset threshold.

7. The method for recognizing the maturity of leaf crops based on small sample images according to claim 1, characterized in that: According to the test group, verify the trained Bayesian model, including: The accuracy of the Bayesian model is tested through the test group. When the accuracy is less than the preset accuracy threshold, the third preset number is updated based on the preset growth rate, and the Bayesian model is obtained again until the accuracy is greater than or equal to the preset accuracy threshold, thereby obtaining a trained Bayesian model.

8. A system for recognizing the maturity of leaf crops based on small sample images, characterized in that: The system comprises: A collection module, used for collecting a second preset number of leaves from the first preset number of crop plants from bottom to top; dividing the leaves into a sample group and a test group; wherein both the sample group and the test group are marked with a preset maturity degree; The union acquisition module is used to extract the color level mean of the Red, Green, Blue channels and grayscale images of the third preset number of leaves at each preset maturity in the sample group to form a table corresponding to each preset maturity; based on the color level mean, the pixel union method is used to perform a union operation on the tables corresponding to each preset maturity to obtain a union data set of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity; The model building module is used to extract the feature interval of the union data set; the feature interval endpoint values ​​are used as training data to train the Bayesian model, and the trained Bayesian model is verified based on the test group.

9. The system for recognizing the maturity of leaf crops based on small sample images according to claim 6, characterized in that: The union acquisition module includes a union unit, The [ ] operator of MATLAB software is used to perform union operations on the color level means of the Red, Green, Blue channels and grayscale images corresponding to each preset maturity, and obtain the corresponding union data sets of the Red, Green, Blue channels and grayscale images at each preset maturity.

Images