Deep learning intelligent detection system and method for multiple pesticide residues of edible agricultural products

Abstract

The invention relates to a deep learning intelligent detection system for multiple pesticide residues of edible agricultural products, and relates to a measuring instrument for various organic matters, in particular to the field of pesticide residue detection.The system comprises a data acquisition device used for acquiring the enzyme inhibition rate of fruit and vegetable agricultural products with fixed volumes for acetylcholin esterase within a set measurement duration; and the intelligent detection device is used for detecting the excessive pesticide residue of the fruit and vegetable agricultural products according to the enzyme inhibition ratio and the customized visual information by using the intelligent detection model for the excessive pesticide residue. The invention also relates to a deep learning intelligent detection method for multiple pesticide residues in edible agricultural products. According to the invention, in order to solve the technical problem that the detection of multiple pesticide residues in agricultural products is difficult to consider the advantages of various organic measurement mechanisms, the intelligent detection of multiple pesticide residue data of fruit and vegetable agricultural products subjected to spot check is completed by using the intelligent detection model for multiple pesticide residues according to different measurement data obtained by the organic matter measurement instruments adopting different measurement principles; therefore, the technical problem is solved.

Description

TECHNICAL FIELD

[0001] The various organic matter measuring instruments of the present application are more particularly related to the field of pesticide residue detection, and in particular to a deep learning intelligent detection system and method for multiple pesticide residues in edible agricultural products. BACKGROUND

[0002] There are various organic matter measuring methods for detecting multiple pesticide residues including organophosphorus pesticide residues and carbamate pesticide residues in fruit and vegetable agricultural products, such as enzyme inhibition rate method which belongs to biochemical assay method, or gas chromatography / mass spectrometry measurement method which belongs to chromatographic detection method. Different organic matter measuring methods have different advantages and disadvantages in measuring accuracy, measuring speed and measuring cost, for example, enzyme inhibition rate method has faster measuring speed but insufficient measuring accuracy. By performing multiple pesticide residue detection on fruit and vegetable agricultural products, it is helpful to understand the degree of pesticide residue in fruit and vegetable agricultural products and to help take subsequent strategies to deal with pesticide residues in fruit and vegetable agricultural products, so as to control the harm of pesticide residues in fruit and vegetable agricultural products of different sources to human body.

[0003] For example, Chinese invention patent publication CN117434228A proposes a pesticide residue detection device for edible agricultural products, which comprises a detection table, support legs are fixedly installed at the bottom of the detection table, a control panel is fixedly installed on the front of the detection table, a placing groove is formed on the front of the detection table, a crushing and liquid discharge mechanism is arranged on the top of the detection table, a liquid separation mechanism is arranged in the placing groove, the crushing and liquid discharge mechanism comprises a crushing unit and a juice separation unit. The connecting sleeve and the crushing blades are crushed and cut on the agricultural products by driving the rotating shaft to rotate by the motor, the connecting sleeve and the rotating shaft can be disassembled, which facilitates cleaning of the connecting sleeve and the crushing blades, the larger agricultural product residues are filtered by the screen plate, the screen plate can be taken out by the lifting rod, which facilitates cleaning of the agricultural product residues on the screen plate, further reducing the cleaning difficulty of the crushing barrel, and the support piece can support the screen plate, facilitating the dripping of the separated agricultural product juice.

[0004] For example, Chinese invention patent publication CN117990619A proposes a pesticide residue detection device for edible agricultural products, which comprises a shell, a display screen fixedly connected in the shell, a cover plate rotatably connected to the surface of the shell, and a detector fixedly connected to the upper surface of the shell. The device further comprises a rotating structure arranged in the shell for continuously detecting samples, the rotating structure comprising a motor fixedly installed on the inner wall of the shell, a contrast frame for rotating the cuvette, and a clearance groove for conveniently taking and placing the cuvette; and a shading structure arranged on the surface of the contrast frame for preventing natural light from affecting the detection accuracy of the detector. During the detection of the pesticide residue concentration of edible agricultural products, the rotating contrast frame can be used to take out and replace a group of cuvettes after detection, thereby improving the detection efficiency.

[0005] However, the above-mentioned various pesticide residue detection schemes for edible agricultural products all use a single detection mechanism to detect pesticide residues in edible agricultural products on site, and cannot take into account the advantages of different pesticide residue detection mechanisms for edible agricultural products, especially cannot use physical data obtained by performing photoelectric imaging on edible agricultural products to detect the degree of pesticide residue of edible agricultural products on site, resulting in detection results that can only show advantages in one of measurement accuracy, measurement speed or measurement cost, and cannot simultaneously take into account the comprehensiveness, accuracy, speed and cost of edible agricultural product pesticide residue detection. SUMMARY

[0006] To solve the technical problems in the prior art, the present application provides a multi-pest residue deep learning intelligent detection system and method for edible agricultural products. For the fixed volume of fruit and vegetable agricultural products selected for inspection, the enzyme inhibition rate of the fruit and vegetable agricultural products and the customized screening visual information of the area occupied by the fruit and vegetable agricultural products in the photoelectric imaging panoramic picture are obtained by using organic matter measuring instruments with different measurement principles. A multi-pest residue intelligent detection model designed with a customized structure is used to complete the intelligent detection of the multi-pest residue data of the fruit and vegetable agricultural products selected for inspection based on the measurement results of different measurement principles. The multi-pest residue data includes organophosphorus pesticide residue and carbamate pesticide residue, thereby taking into account the comprehensiveness, accuracy, speed and cost of batch multi-pest residue detection of fruit and vegetable agricultural products.

[0007] According to an aspect of the present application, a multi-pest residue deep learning intelligent detection system for edible agricultural products is provided, which comprises: A data acquisition device is used to acquire the enzyme inhibition rate of a fixed volume of fruit and vegetable agricultural products against acetylcholinesterase within a set measurement time, and the enzyme inhibition rate is numerically represented by percentage. The information capturing device is used to obtain a panoramic image of the fixed volume of fruits and vegetables before starting the enzyme inhibition rate collection, and the image area occupied by the fruits and vegetables in the panoramic image is taken as the agricultural product capturing area. The continuous learning device is used to continuously perform multiple learning actions on the deep neural network to obtain the deep neural network after completing the multiple learning actions, and output the deep neural network as a multi-residue intelligent detection model, and the number of learning actions performed by the deep neural network is positively correlated with the fixed volume. The intelligent detection device is connected with the data collection device, the information capturing device and the continuous learning device respectively, and is used to intelligently detect the organic phosphorus pesticide residue and the carbamate pesticide residue of the fixed volume of fruits and vegetables according to the fixed volume, the set measurement time, the set imaging distance, the enzyme inhibition rate, the depth of field value, the hue component value, the brightness component value, the saturation component value and the brightness component value gradient of each component pixel point in the agricultural product capturing area by using the multi-residue intelligent detection model. Before starting the enzyme inhibition rate collection, the composite camera mechanism is used to perform the acquisition of the panoramic image of the fixed volume of fruits and vegetables, and the composite camera mechanism is a plurality of cameras which are uniformly arranged around the fixed volume of fruits and vegetables and have a set imaging distance. The brightness component value gradient of each component pixel point is the mean square error of the respective parts of the brightness component value corresponding to each pixel point adjacent to the component pixel point.

[0008] According to another aspect of the present application, a multi-residue deep learning intelligent detection method for edible agricultural products is provided, which comprises: The enzyme inhibition rate of the fixed volume of fruits and vegetables against acetylcholinesterase within a set measurement time is collected, and the enzyme inhibition rate is expressed by a percentage. The panoramic image of the fixed volume of fruits and vegetables before starting the enzyme inhibition rate collection is obtained, and the image area occupied by the fruits and vegetables in the panoramic image is taken as the agricultural product capturing area. The deep neural network is continuously performed multiple learning actions to obtain the deep neural network after completing the multiple learning actions, and output the deep neural network as a multi-residue intelligent detection model, and the number of learning actions performed by the deep neural network is positively correlated with the fixed volume. The multi-residue intelligent detection model is used to intelligently detect the organic phosphorus pesticide residue and the carbamate pesticide residue of the fixed volume of fruits and vegetables according to the fixed volume, the set measurement time, the set imaging distance, the enzyme inhibition rate, the depth of field value, the hue component value, the brightness component value, the saturation component value and the brightness component value gradient of each component pixel point in the agricultural product capturing area. Wherein, before starting the enzyme inhibition rate collection, a composite camera mechanism is used to perform panoramic image acquisition on the fixed volume of fruit and vegetable agricultural products, and the composite camera mechanism is a plurality of cameras that are uniformly arranged around the fixed volume of fruit and vegetable agricultural products and have a distance to the fixed volume of fruit and vegetable agricultural products that is set as an imaging distance. Wherein, the luminance component value gradient of each component pixel point is the mean square error of each part of the luminance component value corresponding to each pixel point adjacent to the component pixel point.

[0009] Therefore, the present application has at least the following four key points: The first is that for the fixed volume of fruit and vegetable agricultural products for sampling inspection, organic matter measuring instruments with different measurement principles are used to acquire the enzyme inhibition rate of the fruit and vegetable agricultural products and the customized screening visual information of the fruit and vegetable agricultural products occupying area in the photoelectric imaging panoramic picture, and an artificial intelligence model is used to complete intelligent detection of multiple pesticide residue data of the fruit and vegetable agricultural products for sampling inspection according to the measurement results of different measurement principles, the multiple pesticide residue data including organophosphorus pesticide residue and carbamate pesticide residue, thereby improving the comprehensiveness and accuracy of batch multiple pesticide residue detection of fruit and vegetable agricultural products; The second is that the artificial intelligence model for intelligent detection of multiple pesticide residue data of the fixed volume of fruit and vegetable agricultural products is a multiple pesticide residue intelligent detection model with customized structure, and the multiple pesticide residue intelligent detection model is a deep neural network after completing multiple learning actions, wherein the number of learning actions performed by the deep neural network is positively correlated with the fixed volume, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, the plurality of hidden layers being located between the single input layer and the single output layer and the number of hidden layers being positively correlated with the fixed volume, and the multiple structure customization designs of the above multiple pesticide residue intelligent detection model ensure the reliability and stability of the intelligent detection results of the multiple pesticide residue data; The third is that the intelligent detection of multiple pesticide residue data of the fixed volume of fruit and vegetable agricultural products is completed by using multiple basic data for targeted screening, the multiple basic data including the enzyme inhibition rate of the fruit and vegetable agricultural products and the customized screening visual information of the fruit and vegetable agricultural products occupying area in the photoelectric imaging panoramic picture, wherein the customized screening visual information of the fruit and vegetable agricultural products occupying area in the photoelectric imaging panoramic picture is the depth of field value, the hue component value, the luminance component value, the saturation component value and the luminance component value gradient of each component pixel point of the fruit and vegetable agricultural products occupying area in the photoelectric imaging panoramic picture, and the targeted screening of the above basic data further ensures the reliability and stability of the intelligent detection results of the multiple pesticide residue data; The fourth place: in each learning action performed on the deep neural network, a certain volume of fruit and vegetable agricultural products that have completed pesticide residue detection is taken as the detected fruit and vegetable agricultural products, the known organophosphorus pesticide residue and carbamate pesticide residue of the detected fruit and vegetable agricultural products are taken as two output contents of the deep neural network, the volume of the detected fruit and vegetable agricultural products, the corresponding measurement time when the enzyme inhibition rate collection is performed, the corresponding imaging distance when the panoramic image picture acquisition is performed, the enzyme inhibition rate collected by performing the enzyme inhibition rate collection, and the depth of field values, hue component values, brightness component values, saturation component values and brightness component value gradients of each constituent pixel point of the corresponding agricultural product capture region are taken as the item-by-item input content of the deep neural network, and the learning action is completed, thereby ensuring the learning effect of each learning action performed on the deep neural network. BRIEF DESCRIPTION OF DRAWINGS

[0010] The embodiments of the present application will be described below in conjunction with the accompanying drawings, in which: Figure 1 The working scene schematic diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system and method according to the present application is shown.

[0011] Figure 2 The internal structure diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system according to the first embodiment of the present application is shown.

[0012] Figure 3 The internal structure diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system according to the second embodiment of the present application is shown.

[0013] Figure 4 The internal structure diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system according to the third embodiment of the present application is shown.

[0014] Figure 5 The internal structure diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system according to the fourth embodiment of the present application is shown.

[0015] Figure 6 The internal structure diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system according to the fifth embodiment of the present application is shown.

[0016] Figure 7 The step flow chart of the edible agricultural product multi-pesticide residue deep learning intelligent detection method according to the sixth embodiment of the present application is shown. DETAILED DESCRIPTION

[0017] As Figure 1 shown, the working scene schematic diagram of the edible agricultural product multi-pesticide residue deep learning intelligent detection system and method according to the present application is shown.

[0018] This invention relates to a deep learning-based intelligent detection system and method for multiple pesticide residues in edible agricultural products, encompassing various organic matter measuring instruments, and more specifically, the field of pesticide residue detection. The specific technical process of this invention is as follows: Technical Process A: To achieve intelligent detection of multiple pesticide residues in a fixed volume of sampled fruits, vegetables and agricultural products, a customized intelligent detection model for multiple pesticide residues is designed. For example, the structural customization of the multi-pesticide residue intelligent detection model is mainly reflected in the following aspects: First: The multi-pesticide residue intelligent detection model is a deep neural network that has completed multiple learning actions. The number of learning actions performed by the deep neural network is positively correlated with the fixed volume, thereby allowing for the customization of multi-pesticide residue intelligent detection models with different structures for fruits and vegetables of different sizes. Second: The deep neural network used includes multiple hidden layers, a single input layer, and a single output layer. The multiple hidden layers are located between the single input layer and the single output layer, and the number of hidden layers is positively correlated with a fixed volume, thereby further customizing and designing intelligent detection models for multiple pesticide residues with different structures for fruits and vegetables of different sizes. Third: In each learning action performed on the deep neural network, a certain volume of fruit and vegetable products that have completed pesticide residue testing is taken as the tested fruit and vegetable products. The known organophosphorus pesticide residues and carbamate pesticide residues of the tested fruit and vegetable products are taken as two output contents of the deep neural network. The volume of the tested fruit and vegetable products, the measurement time corresponding to the enzyme inhibition rate acquisition, the imaging distance corresponding to the panoramic image acquisition, the enzyme inhibition rate acquired by the enzyme inhibition rate acquisition, and the depth, hue, luminance, saturation and luminance gradient values ​​of each constituent pixel in the corresponding agricultural product capture area are taken as the input contents of the deep neural network to complete the learning action. This ensures the learning effect of each learning action performed on the deep neural network. In this way, the reliability and stability of the intelligent detection results of multiple pesticide residues are ensured through the customized structural design of the above-mentioned intelligent detection model. Technical Process B: In order to complete the intelligent detection of multiple pesticide residues in a fixed volume of sampled fruits and vegetables, a number of basic data for targeted screening are introduced, namely, the enzyme inhibition rate of fruits and vegetables and customized screening visual information of the area occupied by fruits and vegetables in the photoelectric imaging panoramic image, which are obtained by organic matter measuring instruments with different measurement principles. like Figure 1 As shown, the various inputs to the multi-pesticide residue intelligent detection model include several types of basic data that are specifically selected. Figure 1enzyme inhibition rate in the enzyme inhibition rate, custom screening visual information and other auxiliary data, including fixed volume, set measurement duration and set imaging distance; Specifically, the enzyme inhibition rate of the fixed volume of fruit and vegetable agricultural products against acetylcholinesterase is collected within the set measurement duration, and the enzyme inhibition rate is expressed by percentage, for example, 50% enzyme inhibition rate; Specifically, the custom screening visual information of the fruit and vegetable agricultural products in the electro-optical imaging panoramic picture is the depth of field value, hue component value, brightness component value, saturation component value and brightness component value gradient of each constituent pixel point of the fruit and vegetable agricultural products in the electro-optical imaging panoramic picture; Further specifically, the brightness component value gradient of each constituent pixel point is the mean square error of the respective brightness component values corresponding to each adjacent pixel point of the constituent pixel point; And further specifically, before starting the enzyme inhibition rate collection of the fixed volume of fruit and vegetable agricultural products, the composite camera mechanism is used to perform panoramic image acquisition on the fixed volume of fruit and vegetable agricultural products, the composite camera mechanism is a plurality of cameras uniformly arranged around the fixed volume of fruit and vegetable agricultural products and having a set imaging distance, and the plurality of cameras have the same resolution; In this way, through the targeted screening of the above basic data, the reliability and stability of the multi-residue data intelligent detection result are further ensured; Technical process C: The multi-residue intelligent detection model designed according to technical process A detects the multi-residue data of the fixed volume of fruit and vegetable agricultural products according to the different physical data obtained by the organic matter measuring instrument with different measurement principles screened according to technical process B; Specifically, the multi-residue data obtained by intelligent detection is the amount of organophosphorus pesticide residues or the amount of carbamate pesticide residues of the fixed volume of fruit and vegetable agricultural products; Technical process D: According to the multi-residue data of the fixed volume of fruit and vegetable agricultural products obtained by intelligent detection according to technical process C, it is determined whether the batch of fruit and vegetable agricultural products where the fixed volume of fruit and vegetable agricultural products is located belongs to the excessive residue batch, and when the batch of fruit and vegetable agricultural products where the fixed volume of fruit and vegetable agricultural products is located belongs to the excessive residue batch, the batch of fruit and vegetable agricultural products where the fixed volume of fruit and vegetable agricultural products is located is reported to the remote residue monitoring server as the excessive residue batch; For example, the remote residue monitoring server is a big data monitoring server, a cloud computing monitoring server or a blockchain monitoring server; Therefore, for a fixed volume of fruits and vegetables sampled for inspection, organic matter measuring instruments with different measurement principles are used to obtain the enzyme inhibition rate of the fruits and vegetables and the customized screening visual information of the area occupied by the fruits and vegetables in the panoramic image of photoelectric imaging. Then, an artificial intelligence model is used to complete the intelligent detection of multiple pesticide residues in the sampled fruits and vegetables based on the measurement results of different measurement principles, thus taking into account the comprehensiveness, accuracy, speed and cost of batch-to-batch multi-pesticide residue detection of fruits and vegetables.

[0019] The key points of this invention are: the introduction of photoelectric physical data for intelligent detection of multiple pesticide residues, the combined use of photoelectric physical data and enzyme inhibition data of edible agricultural products, and the customized structural design of various parts of the intelligent detection model for multiple pesticide residues.

[0020] The following will describe in detail, by way of embodiments, a deep learning intelligent detection system and method for multiple pesticide residues in edible agricultural products according to the present invention. Example

[0021] Figure 2 This is an internal structural diagram of a deep learning intelligent detection system for multiple pesticide residues in edible agricultural products, as shown in the first embodiment of the present invention.

[0022] like Figure 2 As shown, the deep learning intelligent detection system for multiple pesticide residues in edible agricultural products includes the following components: A data acquisition device is used to collect the enzyme inhibition rate of acetylcholinesterase in a fixed volume of fruit and vegetable products within a set measurement time period, wherein the enzyme inhibition rate is expressed as a percentage. Specifically, the enzyme inhibition rate method utilizes the toxicological principle that organophosphates and carbamates can specifically inhibit the activity of acetylcholinesterase in the central and peripheral nervous systems of insects, disrupting normal nerve conduction and causing death by poisoning. Based on this principle, this rapid pesticide residue detection technology uses the inhibitory effect of pesticides on the activity of acetylcholinesterase in vegetable samples to affect the speed of the colorimetric reaction, thereby measuring the pesticide residue (enzyme inhibition rate) and determining whether organophosphates and carbamates in the sample exceed the standard. An information capture device is used to acquire panoramic images of a fixed volume of fruit and vegetable products before the enzyme inhibition rate acquisition is initiated, and the image area occupied by the fruit and vegetable products in the panoramic image is used as the product capture area. For example, acquiring a panoramic image of a fixed volume of fruits and vegetables before the enzyme inhibition rate acquisition is initiated, and using the image area occupied by the fruits and vegetables in the panoramic image as the agricultural product capture area includes: identifying the image area occupied by the fruits and vegetables in the panoramic image based on the imaging characteristics of the fruits and vegetables. Specifically, the identification of the image area occupied by fruits and vegetables in the panoramic image based on the imaging characteristics of fruits and vegetables includes: the imaging characteristics of fruits and vegetables can be the baseline shape outline of fruits and vegetables, or the brightness value range of fruits and vegetables, or the baseline shape outline and the brightness value range of fruits and vegetables can be used simultaneously as the imaging characteristics of fruits and vegetables. A continuous learning device is used to continuously perform multiple learning actions on a deep neural network to obtain a deep neural network after completing multiple learning actions, and output it as a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with a fixed volume. For example, the MATLAB toolbox can be used to perform multiple learning actions on a deep neural network to obtain a deep neural network after multiple learning actions, and then test and simulate the model building process of the output of the multi-pesticide residue intelligent detection model. For example, a deep neural network is continuously subjected to multiple learning actions to obtain a deep neural network after completing multiple learning actions, which is then output as a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with the fixed volume, including: when the fixed volume is 0.1 cubic meters, the number of learning actions performed by the deep neural network is 500; when the fixed volume is 0.2 cubic meters, the number of learning actions performed by the deep neural network is 600; when the fixed volume is 0.5 cubic meters, the number of learning actions performed by the deep neural network is 800; when the fixed volume is 0.8 cubic meters, the number of learning actions performed by the deep neural network is 1000, and so on. The intelligent detection device is connected to the data acquisition device, the information capture device, and the continuous learning device, respectively. It is used to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues of a fixed volume of fruit and vegetable products based on a multi-pesticide residue intelligent detection model, according to a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the gradient values ​​of depth, hue, brightness, saturation, and brightness components of each pixel in the captured area of ​​the agricultural product. Specifically, each constituent pixel has a hue component value, a brightness component value, and a saturation component value in the HSB space, and the value of any component in the hue component value, brightness component value, and saturation component value in the HSB space of each constituent pixel is between 0 and 255. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. For example, before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras may have the same internal structure. The gradient of the luminance component values ​​of each constituent pixel is the mean square error of the luminance component values ​​of each of the neighboring pixels. Specifically, the gradient of the brightness component values ​​of each constituent pixel is the mean square error of the brightness component values ​​of each pixel adjacent to the constituent pixel. The pixels adjacent to the constituent pixel refer to the surrounding pixels that are in contact with the constituent pixel, that is, there are no other pixels between them. The process of continuously performing multiple learning actions on a deep neural network to obtain a deep neural network after completing multiple learning actions, and outputting it as a multi-pesticide residue intelligent detection model, wherein the number of learning actions performed by the deep neural network is positively correlated with a fixed volume, further includes: the deep neural network includes multiple hidden layers, a single input layer and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer; The deep neural network includes multiple hidden layers, a single input layer, and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer. In the deep neural network, the number of hidden layers is positively correlated with a fixed volume. For example, in deep neural networks, the positive correlation between the number of hidden layers and a fixed volume includes: a fixed volume of 0.1 cubic meters corresponds to 3 hidden layers; a fixed volume of 0.2 cubic meters corresponds to 5 hidden layers; a fixed volume of 0.5 cubic meters corresponds to 7 hidden layers; a fixed volume of 0.8 cubic meters corresponds to 9 hidden layers, and so on. In each learning action performed on the deep neural network, a certain volume of fruit and vegetable products that have completed pesticide residue testing is taken as the tested fruit and vegetable products. The known organophosphorus pesticide residues and carbamate pesticide residues of the tested fruit and vegetable products are taken as two output contents of the deep neural network. The volume of the tested fruit and vegetable products, the measurement time corresponding to the enzyme inhibition rate acquisition, the imaging distance corresponding to the panoramic image acquisition, the enzyme inhibition rate acquired by the enzyme inhibition rate acquisition, and the depth, hue, luminance, saturation and luminance gradient values ​​of each constituent pixel in the corresponding agricultural product capture area are taken as the input contents of the deep neural network to complete the learning action. Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras also have the same resolution. For example, multiple cameras having the same resolution includes multiple cameras having an ultra-high definition resolution. Example

[0023] Figure 3 This is an internal structural diagram of a deep learning intelligent detection system for multiple pesticide residues in edible agricultural products, as shown in the second embodiment of the present invention.

[0024] like Figure 3 As shown, compared to Figure 2 The deep learning-based intelligent detection system for multiple pesticide residues in edible agricultural products also includes: The wireless reporting device, connected to the intelligent detection device, is used to wirelessly report the batch of fruits and vegetables containing a fixed volume of fruit and vegetable products as a batch with excessive pesticide residues to a remote pesticide residue monitoring server when the organophosphorus pesticide residue level in a fixed volume of fruit and vegetable products exceeds the organophosphorus pesticide residue threshold and / or the carbamate pesticide residue level in a fixed volume of fruit and vegetable products exceeds the carbamate pesticide residue threshold. Specifically, when the organophosphorus pesticide residue in a fixed volume of fruits and vegetables exceeds the organophosphorus pesticide residue threshold and / or the carbamate pesticide residue in a fixed volume of fruits and vegetables exceeds the carbamate pesticide residue threshold, wirelessly reporting the batch of fruits and vegetables containing the fixed volume of fruits and vegetables as a batch with excessive pesticide residue to a remote pesticide residue monitoring server includes: wirelessly reporting the batch of fruits and vegetables containing the fixed volume of fruits and vegetables as a batch with excessive pesticide residue to a remote pesticide residue monitoring server via a time-division duplex communication link or a frequency-division duplex communication link. Example

[0025] Figure 4 This is an internal structural diagram of a deep learning intelligent detection system for multiple pesticide residues in edible agricultural products, as shown in the third embodiment of the present invention.

[0026] like Figure 4 As shown, compared to Figure 3 The deep learning-based intelligent detection system for multiple pesticide residues in edible agricultural products also includes: The pesticide residue monitoring server is wirelessly connected to the wireless reporting device via a mobile communication link, and is used to receive batches of pesticide residues exceeding the limit wirelessly reported by the wireless reporting device. Among them, the pesticide residue monitoring server is wirelessly connected to the wireless reporting device via a mobile communication link, and is used to receive batches of excessive pesticide residues wirelessly reported by the wireless reporting device. The pesticide residue monitoring server is a big data monitoring server, a cloud computing monitoring server, or a blockchain monitoring server. For example, the pesticide residue monitoring server can be a big data monitoring server, a cloud computing monitoring server, or a blockchain monitoring server, including: the pesticide residue monitoring server can be a big data monitoring network element, a cloud computing monitoring network element, or a blockchain monitoring network element. Example

[0027] Figure 5 This is an internal structural diagram of a deep learning intelligent detection system for multiple pesticide residues in edible agricultural products, as shown in the fourth embodiment of the present invention.

[0028] like Figure 5 As shown, compared to Figure 2 The deep learning-based intelligent detection system for multiple pesticide residues in edible agricultural products also includes: The real-time display device is connected to the intelligent detection device to receive the residue levels of organophosphorus pesticides and carbamate pesticides in a fixed volume of fruits and vegetables, and to perform synchronous real-time display of the residue levels of organophosphorus pesticides and carbamate pesticides in a fixed volume of fruits and vegetables. For example, a real-time display device, connected to an intelligent detection device, is used to receive the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruit and vegetable products, and to perform synchronous real-time display of the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruit and vegetable products. This includes: selecting an LCD screen or an LED screen to implement the real-time display device, connecting it to the intelligent detection device, and using it to receive the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruit and vegetable products, and to perform synchronous real-time display of the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruit and vegetable products. Example

[0029] Figure 6 This is an internal structural diagram of a deep learning intelligent detection system for multiple pesticide residues in edible agricultural products, as shown in the fifth embodiment of the present invention.

[0030] like Figure 6 As shown, compared toFigure 2 The deep learning-based intelligent detection system for multiple pesticide residues in edible agricultural products also includes: The model storage device, connected to the continuous learning device, is used to store the model of the multi-pesticide residue intelligent detection model by storing various model parameters.

[0031] For example, a model storage device, connected to a continuous learning device, is used to store the model of a multi-pesticide residue intelligent detection model by storing various model parameters of the model. This includes: the model storage device can be implemented using an MMC storage device or a FLASH storage device, connected to the continuous learning device, and used to store the model of a multi-pesticide residue intelligent detection model by storing various model parameters of the model.

[0032] Next, various embodiments of the present invention will be further described.

[0033] Optionally, within the above embodiments, in the deep learning intelligent detection system for multiple pesticide residues in edible agricultural products: The multi-pesticide residue intelligent detection model is used to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables based on a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the depth, hue, luminance, saturation, and luminance component values ​​of each pixel in the captured area of ​​the agricultural product. The method involves performing numerical normalization on the fixed volume, set measurement time, set imaging distance, enzyme inhibition rate, and the depth, hue, luminance, saturation, and luminance component values ​​of each pixel in the captured area of ​​the agricultural product before synchronously inputting them into the multi-pesticide residue intelligent detection model. For example, after performing numerical normalization on the fixed volume, set measurement time, set imaging distance, enzyme inhibition rate, and the depth, hue, brightness, saturation and brightness gradient values ​​of each constituent pixel in the agricultural product capture area, the data are then synchronously input into the multi-pesticide residue intelligent detection model. The setting of the imaging distance can be 0.5 meters. The method of using a multi-pesticide residue intelligent detection model to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables based on a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the gradient values ​​of depth, hue, brightness, saturation, and brightness components of each pixel in the captured area of ​​the agricultural product. This method also includes: running the multi-pesticide residue intelligent detection model to obtain the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables in a normalized representation of the values ​​output by the multi-pesticide residue intelligent detection model. Among them, the deep neural network is continuously executed multiple learning actions to obtain a deep neural network after completing multiple learning actions, and is used as the output of the multi-pesticide residue intelligent detection model. The positive correlation between the number of learning actions executed by the deep neural network and the fixed volume includes: using a numerical mapping function to represent the numerical mapping relationship between the number of learning actions executed by the deep neural network and the fixed volume. The numerical mapping relationship between the number of learning actions performed by the deep neural network and the fixed volume, represented by a numerical mapping function, includes: in the numerical mapping function, the fixed volume is the input value of the numerical mapping function, and the number of learning actions performed by the deep neural network that is positively correlated with the fixed volume is the output value of the numerical mapping function.

[0034] Optionally, within the above embodiments, in the deep learning intelligent detection system for multiple pesticide residues in edible agricultural products: Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras have the same imaging lens, flexible circuit board, photoelectric sensor, power supply circuit, and filter. Specifically, the plurality of cameras have the same imaging lens, flexible circuit board, photoelectric sensor, power supply circuit and filter, including: each camera also has the same type of optical anti-vibration unit, which is disposed on the imaging lens; Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. In each camera, a filter is set between the imaging lens and the photoelectric sensor. In addition, before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, with each camera at a set imaging distance. In each camera, a photoelectric sensor and a power supply circuit are installed on a flexible circuit board, and the photoelectric sensor is a CMOS sensor or a CCD sensor. Example

[0035] Figure 7 This is a flowchart illustrating the steps of a deep learning-based intelligent detection method for multiple pesticide residues in edible agricultural products according to the sixth embodiment of the present invention.

[0036] like Figure 7As shown, the deep learning-based intelligent detection method for multiple pesticide residues in edible agricultural products includes the following steps: Step 71: Data acquisition device, used to collect the enzyme inhibition rate of a fixed volume of fruit and vegetable products against acetylcholinesterase within a set measurement time period, wherein the enzyme inhibition rate is expressed as a percentage. Specifically, the enzyme inhibition rate method utilizes the toxicological principle that organophosphates and carbamates can specifically inhibit the activity of acetylcholinesterase in the central and peripheral nervous systems of insects, disrupting normal nerve conduction and causing death by poisoning. Based on this principle, this rapid pesticide residue detection technology uses the inhibitory effect of pesticides on the activity of acetylcholinesterase in vegetable samples to affect the speed of the colorimetric reaction, thereby measuring the pesticide residue (enzyme inhibition rate) and determining whether organophosphates and carbamates in the sample exceed the standard. Step 72: Obtain a panoramic image of a fixed volume of fruits and vegetables before the enzyme inhibition rate acquisition is initiated, and use the image area occupied by the fruits and vegetables in the panoramic image as the agricultural product capture area. For example, acquiring a panoramic image of a fixed volume of fruits and vegetables before the enzyme inhibition rate acquisition is initiated, and using the image area occupied by the fruits and vegetables in the panoramic image as the agricultural product capture area includes: identifying the image area occupied by the fruits and vegetables in the panoramic image based on the imaging characteristics of the fruits and vegetables. Specifically, the identification of the image area occupied by fruits and vegetables in the panoramic image based on the imaging characteristics of fruits and vegetables includes: the imaging characteristics of fruits and vegetables can be the baseline shape outline of fruits and vegetables, or the brightness value range of fruits and vegetables, or the baseline shape outline and the brightness value range of fruits and vegetables can be used simultaneously as the imaging characteristics of fruits and vegetables. Step 73: Execute multiple learning actions on the deep neural network to obtain the deep neural network after completing multiple learning actions, and output it as the multi-pesticide residue intelligent detection model. The number of learning actions executed by the deep neural network is positively correlated with the fixed volume. For example, the MATLAB toolbox can be used to perform multiple learning actions on a deep neural network to obtain a deep neural network after multiple learning actions, and then test and simulate the model building process of the output of the multi-pesticide residue intelligent detection model. For example, a deep neural network is continuously subjected to multiple learning actions to obtain a deep neural network after completing multiple learning actions, which is then output as a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with the fixed volume, including: when the fixed volume is 0.1 cubic meters, the number of learning actions performed by the deep neural network is 500; when the fixed volume is 0.2 cubic meters, the number of learning actions performed by the deep neural network is 600; when the fixed volume is 0.5 cubic meters, the number of learning actions performed by the deep neural network is 800; when the fixed volume is 0.8 cubic meters, the number of learning actions performed by the deep neural network is 1000, and so on. Step 74: Using a multi-pesticide residue intelligent detection model, based on a fixed volume, set measurement time, set imaging distance, enzyme inhibition rate, and the gradient values ​​of depth of field, hue component, brightness component, saturation component, and brightness component of each constituent pixel in the agricultural product capture area, intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruit and vegetable agricultural products. Specifically, each constituent pixel has a hue component value, a brightness component value, and a saturation component value in the HSB space, and the value of any component in the hue component value, brightness component value, and saturation component value in the HSB space of each constituent pixel is between 0 and 255. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. For example, before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras may have the same internal structure. The gradient of the luminance component values ​​of each constituent pixel is the mean square error of the luminance component values ​​of each of the neighboring pixels. Specifically, the gradient of the brightness component values ​​of each constituent pixel is the mean square error of the brightness component values ​​of each pixel adjacent to the constituent pixel. The pixels adjacent to the constituent pixel refer to the surrounding pixels that are in contact with the constituent pixel, that is, there are no other pixels between them. The process of continuously performing multiple learning actions on a deep neural network to obtain a deep neural network after completing multiple learning actions, and outputting it as a multi-pesticide residue intelligent detection model, wherein the number of learning actions performed by the deep neural network is positively correlated with a fixed volume, further includes: the deep neural network includes multiple hidden layers, a single input layer and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer; The deep neural network includes multiple hidden layers, a single input layer, and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer. In the deep neural network, the number of hidden layers is positively correlated with a fixed volume. For example, in deep neural networks, the positive correlation between the number of hidden layers and a fixed volume includes: a fixed volume of 0.1 cubic meters corresponds to 3 hidden layers; a fixed volume of 0.2 cubic meters corresponds to 5 hidden layers; a fixed volume of 0.5 cubic meters corresponds to 7 hidden layers; a fixed volume of 0.8 cubic meters corresponds to 9 hidden layers, and so on. In each learning action performed on the deep neural network, a certain volume of fruit and vegetable products that have completed pesticide residue testing is taken as the tested fruit and vegetable products. The known organophosphorus pesticide residues and carbamate pesticide residues of the tested fruit and vegetable products are taken as two output contents of the deep neural network. The volume of the tested fruit and vegetable products, the measurement time corresponding to the enzyme inhibition rate acquisition, the imaging distance corresponding to the panoramic image acquisition, the enzyme inhibition rate acquired by the enzyme inhibition rate acquisition, and the depth, hue, luminance, saturation and luminance gradient values ​​of each constituent pixel in the corresponding agricultural product capture area are taken as the input contents of the deep neural network to complete the learning action. Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras also have the same resolution. For example, multiple cameras having the same resolution includes multiple cameras having an ultra-high definition resolution.

[0037] Furthermore, in the deep learning intelligent detection system and method for multiple pesticide residues in edible agricultural products according to the present invention: The process involves performing numerical normalization on the fixed volume, measurement duration, imaging distance, enzyme inhibition rate, and the depth, hue, brightness, saturation, and brightness gradient values ​​of each pixel in the agricultural product capture area, and then synchronously inputting them into the multi-pesticide residue intelligent detection model. This includes using a first programmable logic device and a second programmable logic device to perform numerical normalization and synchronous input, with the first programmable logic device connected to the second programmable logic device. For example, a first programmable logic device and a second programmable logic device are used to perform numerical normalization processing and synchronous input, respectively. The connection between the first programmable logic device and the second programmable logic device includes: the first programmable logic device can be selected as a CPLD chip, and the second programmable logic device can be selected as an FPGA chip. Both the CPLD chip and the FPGA chip are programmed using VHDL language. Furthermore, the process of performing numerical normalization on the fixed volume, measurement duration, imaging distance, enzyme inhibition rate, and the gradient of depth, hue, brightness, saturation, and brightness components of each pixel in the agricultural product capture area before synchronously inputting them into the multi-pesticide residue intelligent detection model also includes: numerical normalization to octal conversion.

[0038] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0039] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device to execute the methods described in the various embodiments or some parts of the embodiments.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A deep learning-based intelligent detection system for multiple pesticide residues in edible agricultural products, characterized in that, The system includes: A data acquisition device is used to collect the enzyme inhibition rate of acetylcholinesterase in a fixed volume of fruit and vegetable products within a set measurement time period, wherein the enzyme inhibition rate is expressed as a percentage. An information capture device is used to acquire panoramic images of a fixed volume of fruit and vegetable products before the enzyme inhibition rate acquisition is initiated, and the image area occupied by the fruit and vegetable products in the panoramic image is used as the product capture area. A continuous learning device is used to continuously perform multiple learning actions on a deep neural network to obtain a deep neural network after completing multiple learning actions, and output it as a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with a fixed volume. The intelligent detection device is connected to the data acquisition device, the information capture device, and the continuous learning device, respectively. It is used to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues of a fixed volume of fruit and vegetable products based on a multi-pesticide residue intelligent detection model, according to a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the gradient values ​​of depth, hue, brightness, saturation, and brightness components of each pixel in the captured area of ​​the agricultural product. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The gradient of the luminance component values ​​of each constituent pixel is the mean square error of the luminance component values ​​of each of the neighboring pixels.

2. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 1, characterized in that: The deep neural network is subjected to multiple learning actions in succession to obtain a deep neural network after completing multiple learning actions, and is output as a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with a fixed volume. The deep neural network includes multiple hidden layers, a single input layer and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer. The deep neural network includes multiple hidden layers, a single input layer, and a single output layer, wherein the multiple hidden layers are located between the single input layer and the single output layer. In the deep neural network, the number of hidden layers is positively correlated with a fixed volume.

3. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 2, characterized in that: In each learning action performed on the deep neural network, a certain volume of fruit and vegetable products that have completed pesticide residue testing is taken as the tested fruit and vegetable products. The known organophosphorus pesticide residues and carbamate pesticide residues of the tested fruit and vegetable products are taken as two output contents of the deep neural network. The volume of the tested fruit and vegetable products, the measurement time corresponding to the enzyme inhibition rate acquisition, the imaging distance corresponding to the panoramic image acquisition, the enzyme inhibition rate acquired by the enzyme inhibition rate acquisition, and the depth, hue, luminance, saturation and luminance gradient values ​​of each constituent pixel in the corresponding agricultural product capture area are taken as the input contents of the deep neural network to complete the learning action. Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras also include cameras with the same resolution.

4. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 3, characterized in that, The system also includes: The wireless reporting device, connected to the intelligent detection device, is used to wirelessly report the batch of fruits and vegetables containing a fixed volume of fruit and vegetable products as a batch with excessive pesticide residues to a remote pesticide residue monitoring server when the organophosphorus pesticide residue level in a fixed volume of fruit and vegetable products exceeds the organophosphorus pesticide residue threshold and / or the carbamate pesticide residue level in a fixed volume of fruit and vegetable products exceeds the carbamate pesticide residue threshold.

5. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 4, characterized in that, The system also includes: The pesticide residue monitoring server is wirelessly connected to the wireless reporting device via a mobile communication link, and is used to receive batches of pesticide residues exceeding the limit wirelessly reported by the wireless reporting device. Among them, the pesticide residue monitoring server is wirelessly connected to the wireless reporting device via a mobile communication link, and is used to receive batches of excessive pesticide residues wirelessly reported by the wireless reporting device. The pesticide residue monitoring server is a big data monitoring server, a cloud computing monitoring server, or a blockchain monitoring server.

6. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 3, characterized in that, The system also includes: The real-time display device, connected to the intelligent detection device, is used to receive the residue levels of organophosphorus pesticides and carbamate pesticides in a fixed volume of fruits and vegetables, and to perform synchronous real-time display of the residue levels of organophosphorus pesticides and carbamate pesticides in a fixed volume of fruits and vegetables.

7. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in claim 3, characterized in that, The system also includes: The model storage device, connected to the continuous learning device, is used to store the model of the multi-pesticide residue intelligent detection model by storing various model parameters.

8. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in any one of claims 3-7, characterized in that: The multi-pesticide residue intelligent detection model is used to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables based on a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the depth, hue, luminance, saturation, and luminance component values ​​of each pixel in the captured area of ​​the agricultural product. The method involves performing numerical normalization on the fixed volume, set measurement time, set imaging distance, enzyme inhibition rate, and the depth, hue, luminance, saturation, and luminance component values ​​of each pixel in the captured area of ​​the agricultural product before synchronously inputting them into the multi-pesticide residue intelligent detection model. The method of using a multi-pesticide residue intelligent detection model to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables based on a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the gradient values ​​of depth, hue, brightness, saturation, and brightness components of each pixel in the captured area of ​​the agricultural product. This method also includes: running the multi-pesticide residue intelligent detection model to obtain the organophosphorus pesticide residues and carbamate pesticide residues in a fixed volume of fruits and vegetables in a normalized representation of the values ​​output by the multi-pesticide residue intelligent detection model. Among them, the deep neural network is continuously executed multiple learning actions to obtain a deep neural network after completing multiple learning actions, and is used as the output of the multi-pesticide residue intelligent detection model. The positive correlation between the number of learning actions executed by the deep neural network and the fixed volume includes: using a numerical mapping function to represent the numerical mapping relationship between the number of learning actions executed by the deep neural network and the fixed volume. The numerical mapping relationship between the number of learning actions performed by the deep neural network and the fixed volume, represented by the numerical mapping function, includes: in the numerical mapping function, the fixed volume is the input value of the numerical mapping function, and the number of learning actions performed by the deep neural network that is positively correlated with the fixed volume is the output value of the numerical mapping function.

9. The deep learning intelligent detection system for multiple pesticide residues in edible agricultural products as described in any one of claims 3-7, characterized in that: Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The multiple cameras have the same imaging lens, flexible circuit board, photoelectric sensor, power supply circuit, and filter. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. In each camera, a filter is set between the imaging lens and the photoelectric sensor. Before initiating enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. In each camera, a photoelectric sensor and a power supply circuit are installed on a flexible circuit board, and the photoelectric sensor is either a CMOS sensor or a CCD sensor.

10. A deep learning-based intelligent detection method for multiple pesticide residues in edible agricultural products, characterized in that, The method includes: The enzyme inhibition rate of acetylcholinesterase was collected from a fixed volume of fruits and vegetables within a set measurement time period, and the enzyme inhibition rate was expressed as a percentage. A panoramic image of a fixed volume of fruits and vegetables was acquired before the enzyme inhibition rate was collected, and the image area occupied by the fruits and vegetables in the panoramic image was used as the agricultural product capture area. The deep neural network is continuously subjected to multiple learning actions to obtain a deep neural network after completing multiple learning actions, and this network is used as the output of a multi-pesticide residue intelligent detection model. The number of learning actions performed by the deep neural network is positively correlated with the fixed volume. The multi-pesticide residue intelligent detection model is used to intelligently detect the organophosphorus pesticide residues and carbamate pesticide residues of a fixed volume of fruit and vegetable products based on a fixed volume, a set measurement time, a set imaging distance, an enzyme inhibition rate, and the gradient values ​​of depth of field, hue component, brightness component, saturation component, and brightness component of each constituent pixel in the captured area of ​​the agricultural product. Before starting the enzyme inhibition rate acquisition, a composite camera mechanism is used to acquire panoramic images of a fixed volume of fruit and vegetable products. The composite camera mechanism consists of multiple cameras evenly arranged around the fixed volume of fruit and vegetable products, each at a set imaging distance. The gradient of the luminance component values ​​of each constituent pixel is the mean square error of the luminance component values ​​of each of the neighboring pixels.

Images