In-situ soil living small animal layering detection investigation method and device

Abstract

The application discloses an in-situ soil living small animal layered detection investigation method and device, adopts an in-situ soil living small animal layered instance segmentation device, automatically collects multi-modal hyperspectral imaging data of layered soil samples, and utilizes a soil animal 3D convolution instance segmentation deep network model to perform pixel instance segmentation on the multi-modal hyperspectral imaging data, independently identifies the types and instance individual numbers of small animals in each layer of soil, and further calculates the types, instance individual numbers, densities of small animals in the in-situ overall soil sample, and the longitudinal distribution of small animals in the soil. The application breaks through the limitations of traditional soil ecological research, avoids the low efficiency and large error of mechanical soil digging, manual selection and insect trapping and other methods, and provides a fast and accurate soil animal detection technology. The technology provides reliable data support for soil ecological research and agricultural management, and is an advanced tool for soil health management and sustainable agricultural development.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of soil ecology investigation, and particularly relates to a method and device for in-situ soil living small animal layered detection investigation. BACKGROUND

[0002] Soil small animals play a crucial role in ecosystems by decomposing organic matter, promoting nutrient cycling, and improving soil structure. Their activities, such as burrowing and foraging, increase soil porosity and water retention, creating a heterogeneous environment that supports diverse microbial communities. These processes not only promote plant growth but also maintain soil health, making soil small animals essential for sustainable ecosystems. Field investigation of soil small animals is an important part of ecological research. However, traditional investigation methods, such as direct observation and trapping, often require removing soil samples from their natural environment, which is not only time-consuming and labor-intensive but also prone to inaccuracy due to human error and subjective judgment. Moreover, these methods often lack in-depth analysis of soil layering, making it difficult to provide information on the distribution of living soil small animals in different soil depth layers. Although biomarker analysis can provide quantitative data, its complex chemical processing and analysis procedures not only increase costs but also prolong research cycles. SUMMARY

[0003] To overcome the problems in the prior art, the present application proposes a method and device for in-situ soil living small animal layered detection investigation. This method directly detects in-situ soil samples, avoiding interference and contamination during sample transfer, ensuring the authenticity and reliability of the data. At the same time, the layered detection technology of the present application can accurately identify the species, quantity, and distribution of soil small animals in different soil layers, providing a new perspective for the study of soil ecological structure. Compared with traditional methods, the real-time recognition capability of the present application significantly improves the detection efficiency, and the automated soil layering detection method simplifies the operation process and reduces the technical threshold. In addition, the detection method of the present application does not require chemical processing, reducing the impact on the environment and organisms. The present application aims to achieve rapid, accurate, and layered detection of soil living small animals through innovative technology, reduce interference with the soil environment, improve detection efficiency, and reduce costs, providing more reliable data support for soil ecological research and agricultural management. This method provides a more advanced tool for soil ecological research and agricultural practice, promoting soil health management and sustainable agricultural development.

[0004] The technical solutions for achieving the purposes of the present application are as follows:

[0005] A kind of in-situ soil living small animal layered detection investigation method and device, the soil sample collected in nature is placed in in-situ soil living small animal layered instance segmentation device, gas anesthesia system controls anesthesia gas to enter soil sample layered detection box at uniform speed, after small animal in soil sample is narcotized, information processing unit in in-situ soil living small animal layered instance segmentation device controls hyperspectral imaging module to combine broadband light source illumination module to collect the reflection spectrum of small animal in soil sample;Information processing unit controls hyperspectral imaging module to combine ultraviolet light excitation module to detect the fluorescence spectrum of small animal in soil sample;Reflection spectrum and fluorescence spectrum constitute multimodal hyperspectral imaging data, which is subjected to pixel instance segmentation by soil animal 3D convolution instance segmentation deep network model, which identifies the pixel label category of soil small animal while segmenting the pixel region of soil small animal and soil;

[0006] The in-situ soil living small animal layered instance segmentation device adopts layered detection, and the soil animal 3D convolution instance segmentation deep network model independently identifies the species and instance individual number of small animals in each layer of soil, and finally calculates the species, instance individual number, distribution density of small animals in the in-situ overall soil sample, and the longitudinal distribution of small animals in soil.

[0007] The in-situ soil living small animal layered instance segmentation device comprises an instrument light-shielding airtight cover, a broadband light source illumination module, a hyperspectral imaging module, a first electric displacement table, an electric push arm module, a gas anesthesia system, a soil sample layered detection box, a soil sample storage box, an ultraviolet light excitation module, an information processing unit, and a turnable transparent window module.

[0008] The broadband light source illumination module comprises a first wide spectrum light source, a second wide spectrum light source, a first condenser lens, and a second condenser lens.

[0009] The hyperspectral imaging module comprises a first lens, a slit, a second lens, a spectral transformation module, a third lens, a camera, and a second electric displacement platform.

[0010] The electric push arm module comprises a first electric push arm, a second electric push arm, a third electric push arm, a fourth electric push arm, a fifth electric push arm, a sixth electric push arm, and a seventh electric push arm.

[0011] The ultraviolet light excitation module comprises a first ultraviolet light source, a fourth lens, a first rotary motor, a first band-pass filter set, a second ultraviolet light source, a fifth lens, a second rotary motor, a second band-pass filter set, a third ultraviolet light source, a sixth lens, a third rotary motor, a third band-pass filter set, a fourth ultraviolet light source, a seventh lens, a fourth rotary motor, a fourth band-pass filter set, a fifth ultraviolet light source, an eighth lens, a fifth rotary motor, a fifth band-pass filter set, a sixth ultraviolet light source, a ninth lens, a sixth rotary motor, a sixth band-pass filter set, a seventh ultraviolet light source, a tenth lens, a seventh rotary motor, and a seventh band-pass filter set; the first ultraviolet light source, the fourth lens, the first rotary motor, and the first band-pass filter set are connected; the second ultraviolet light source, the fifth lens, the second rotary motor, and the second band-pass filter set are connected; the third ultraviolet light source, the sixth lens, the third rotary motor, and the third band-pass filter set are connected; the fourth ultraviolet light source, the seventh lens, the fourth rotary motor, and the fourth band-pass filter set are connected; the fifth ultraviolet light source, the eighth lens, the fifth rotary motor, and the fifth band-pass filter set are connected; the sixth ultraviolet light source, the ninth lens, the sixth rotary motor, and the sixth band-pass filter set are connected; and the seventh ultraviolet light source, the tenth lens, the seventh rotary motor, and the seventh band-pass filter set are connected.

[0012] The foldable transparent window module comprises a first foldable transparent window, a second foldable transparent window, a third foldable transparent window, a fourth foldable transparent window, a fifth foldable transparent window, a sixth foldable transparent window, and a seventh foldable transparent window; the first foldable transparent window, the second foldable transparent window, the third foldable transparent window, the fourth foldable transparent window, the fifth foldable transparent window, the sixth foldable transparent window, and the seventh foldable transparent window are connected from top to bottom; the information processing unit comprises an industrial computer and a data connection line; the information processing unit is connected with the broadband light source illumination module, the hyperspectral imaging module, the electric push arm module, the gas anesthesia system, the ultraviolet light excitation module, and the first electric displacement table; the information processing unit is used to control the operation timing of the broadband light source illumination module, the ultraviolet light excitation module, the hyperspectral imaging module, the electric push arm module, the gas anesthesia system, and the first electric displacement table, so that the hyperspectral imaging module can orderly collect the multi-modal hyperspectral data of all the layered soil samples in the soil sample layered detection box, and perform instance segmentation on the multi-modal hyperspectral data of the soil sample to calculate the information such as the type, the instance individual quantity, and the distribution density of the small animals in the soil sample.

[0013] The layered detection method, the soil animal 3D convolution instance segmentation deep network model calculates the species and instance number of small animals in the soil surface layer according to the multi-modal hyperspectral imaging data composed of the reflection spectrum and the fluorescence spectrum collected by the in-situ soil living small animal layered instance segmentation device, and estimates the species and instance number of small animals in the deep layer of the soil by using the following method:

[0014] The volume of the soil sample obtained from the soil in-situ is (x, y, z), and the number of small animals of various species in the soil is S (S1, S2, S 3, …,S n ), wherein n is the number of species of small animals in the soil;

[0015] The in-situ soil living small animal layered instance segmentation device divides the soil sample into k layers in the longitudinal position; the current soil sample layered sequence number is q, the volume of each layer of soil sample is (x, y, z / k), and the number of small animals of various species in the qth layer of soil sample is P q (P q 1,P q 2,P q 3, …,P q m ), wherein m is the total number of species of small animals in the qth layer of soil sample;

[0016] The calibration is established as follows: the multi-modal hyperspectral imaging data is composed of the reflection spectrum and the fluorescence spectrum of the surface layer of the first layer of soil sample in the soil sample layered detection box 10 collected by the in-situ soil living small animal layered instance segmentation device, the species and instance number of small animals in the surface layer of the first layer of soil sample are identified by the soil animal 3D convolution instance segmentation deep network model, the species number set T 1 (T 1 1,T 1 2,T 1 3, …,T 1 m ) of the current soil sample surface layer small animals is obtained; the total small animal species number set P 1 (P 1 1,P 1 2,P 1 3, …,P 1 m ) of the first layer of soil sample in the soil sample layered detection box 10 is obtained by manually separating small animals inside the soil sample, selecting and manually identifying; the P 1 (P 1 1,P 1 2,P 1 3, …,P1 m ) and T 1 (T 1 1,T 1 2,T 1 3, …,T 1 m ) is as follows:

[0017]

[0018] According to the above fitting relationship, the dataset T of small animals on the surface of the qth layer of soil samples is obtained by combining the in-situ soil living small animal layer instance segmentation device with the soil animal 3D convolutional instance segmentation deep network model. q (T q 1,T q 2,T q 3, …,T q m ), the total small animal species dataset P of the qth layer soil sample can be calculated q (P q 1,P q 2,P q 3, …,P q m );

[0019] Then, in the soil sample stratification detection box, the distribution density of soil small animals of species i in the qth layer of soil sample is:

[0020]

[0021] Then, in the soil sample stratification detection box, the distribution density of soil small animals in the qth layer of soil sample is:

[0022]

[0023] The above steps are repeated in sequence until all the small animal species data sets of the stratified soil samples in the soil sample stratification detection box are detected;

[0024] The number of individuals of various types of soil animals in the soil sample stratification detection box is:

[0025]

[0026] The total number of soil microfauna in the soil sample stratification detection box is:

[0027]

[0028] The distribution density of various soil animal species in the soil sample stratification test box is as follows:

[0029]

[0030] The total distribution density of soil microfauna is:

[0031]

[0032] The soil animal 3D convolutional instance segmentation deep network model can extract the spatial and spectral features of the multimodal hyperspectral imaging data collected by the in-situ soil living small animal layered instance segmentation device based on the 3D convolutional neural network, so as to perform instance segmentation of small animals in the soil. The species and individual information of the instance segmentation can be used to analyze the growth cycle, male and female individual information of the soil small animal individual, so as to establish a better soil small animal library. Even if the small animal is inside the soil, the soil small animal individuals actually collected are incomplete due to the occlusion of the soil. The soil animal 3D convolutional instance segmentation deep network model defined in the present invention can be used to perform instance segmentation on the incomplete soil small animals, which can not only segment and identify the species and instance individuals of the small animals, but also analyze whether the incomplete small animal parts are covered living small animals or dead bodies with broken limbs.

[0033] In the described stratified detection, when the hyperspectral imaging module in the in-situ soil living small animal stratified instance segmentation device detects the instance segmented species of the first layer of soil small animals, the first electric push arm in the electric push arm module will push the first layer of soil through the first flip-able transparent window into the soil sample storage box; the first electric displacement platform in the hyperspectral imaging module drives the hyperspectral imaging module to move downward so that the surface image of the second layer of soil is clear; after the hyperspectral imaging module detects the instance segmented species of the second layer of soil small animals, the second electric push arm in the electric push arm module pushes the second layer of soil through the second flip-able transparent window into the soil sample storage box; the cycle is repeated until all stratified soil samples in the soil sample stratification detection box are detected.

[0034] In the multimodal hyperspectral imaging detection, a broadband light source illumination module uses a broad spectrum of visible and near-infrared wavelengths (400-1800 nm) to detect soil microfauna. After detecting the reflectance spectrum of the soil sample, the information processing center controls the power supply of the broadband light source illumination module to be turned off and the power supply of the ultraviolet light excitation module to be turned on. When detecting the first layer of soil samples in the soil sample stratification detection box, ultraviolet light emitted by the first ultraviolet light source is collimated by a fourth lens, passes through a first bandpass filter set, and then irradiates the surface of the soil microfauna from the side. The soil microfauna are excited by the ultraviolet light and emit fluorescence, and the hyperspectral imaging module collects fluorescence spectrum data. The first rotating motor drives the first bandpass filter set to rotate to increase the switching of different bandpass filters to achieve different fluorescence wavelength bands. When detecting the next layer of soil samples in the soil sample stratification detection box, the broadband light source illumination process is repeated and the second ultraviolet light source is switched to allow ultraviolet light to irradiate the next layer of soil samples. This cycle is repeated until the multimodal hyperspectral imaging detection of all stratified soil samples in the soil sample stratification detection box is completed.

[0035] In the described hyperspectral imaging module, the reflection or fluorescence image of the surface soil sample in the soil sample stratification detection box is imaged on the slit surface after passing through the first lens, and the light passing through the center of the slit is collimated into parallel light of different angles after passing through the second lens; the parallel light of different angles passes through the spectrum conversion module and then through the third lens to focus on the camera to form a spectral image; the spectrum conversion module is composed of a prism-grating-prism module, or a tunable filter, which is used to separate the complex light into monochromatic light of different bands; the second electric displacement platform drives the hyperspectral imaging module to perform one-dimensional scanning, thereby realizing reflection-fluorescence multimodal hyperspectral imaging detection of the surface soil sample in the soil sample stratification detection box.

[0036] Beneficial effects of the present invention:

[0037] This invention discloses a method and apparatus for in-situ stratified detection and investigation of living soil small animals. Using an in-situ stratified instance segmentation device for living soil small animals, this method can perform stratified multimodal hyperspectral imaging of living soil small animals. Multimodal hyperspectral data is segmented pixel by pixel using a 3D convolutional instance segmentation deep network model for soil animals. Based on the instance segmentation results, the species and number of small animals within each soil layer are independently identified. Finally, the species, density, and number of small animals within the entire in-situ soil sample are calculated.

[0038] The present application overcomes the low efficiency, time-consuming and labor-intensive, and large statistical error and low timeliness caused by the insect attracting method in the traditional soil ecology research by using mechanical soil excavation, screening out impurities, and manually selecting small animals in the soil for empirical identification and classification. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a schematic diagram of a method for in-situ soil living small animal layered detection and investigation.

[0040] Figure 2 It is an in-situ soil living small animal layered example segmentation device.

[0041] Figure 2 In the instrument light-shielded airtight cover 1, the first wide spectrum light source 2, the second wide spectrum light source 3, the first condenser lens 4, the second condenser lens 5, the wideband light source illumination module 6, the hyperspectral imaging module 7, the electric push arm module 8, the gas anesthesia system 9, the soil sample layering detection box 10, the soil sample storage box 11, the first electric displacement table 12, the first lens 13, the slit 14, the second lens 15, the spectral transformation module 16, the third lens 17, the camera 18, the second electric displacement platform 19, the ultraviolet excitation module 20, the first ultraviolet light source 21, the fourth lens 22, the first rotary motor 23, the first band-pass filter set 24, the second ultraviolet light source 25, the fifth lens 26, the second rotary motor 27, the second band-pass filter set 28, the third ultraviolet light source 29, the sixth lens 30, the third rotary motor 31, the third band-pass filter set 32, the fourth ultraviolet light source 33, the seventh lens 34, the fourth rotary motor 35, the fourth band-pass filter set 36, the fifth ultraviolet light source 37, the eighth lens 38, the fifth rotary motor 39, the fifth band-pass filter set 40, the sixth ultraviolet light source 41, the ninth lens 42, the sixth rotary motor 43, the sixth band-pass filter set 44, the seventh ultraviolet light source 45, the tenth lens 46, the seventh rotary motor 47, the seventh band-pass filter set 48, the first electric push arm 49, the second electric push arm 50, the third electric push arm 51, the fourth electric push arm 52, the fifth electric push arm 53, the sixth electric push arm 54, the seventh electric push arm 55, the first flipable transparent window 56, the second flipable transparent window 57, the third flipable transparent window 58, the fourth flipable transparent window 59, the fifth flipable transparent window 60, the sixth flipable transparent window 61, the seventh flipable transparent window 62, the information processing unit 63, and the flipable transparent window module 64.

[0042] Figure 3is an example segmentation device for in-situ soil living small animals, and an example segmentation effect diagram of soil small animals;

[0043] Figure 3 In the figure, the first row from left to right is: earthworm 70, earthworm 71, crustacean 72, whip scorpion 73, ground centipede 74, stone centipede 75, millipede 76, cricket 77. DETAILED DESCRIPTION

[0044] The application will be further described below in conjunction with the drawings and examples.

[0045] An in-situ soil living small animal layering detection method, the flow is as shown in the figure. Figure 1 The natural soil sample with a volume of (x, y, z) is placed in the in-situ soil living small animal layering example segmentation device, the gas anesthesia system controls the anesthesia gas to pass into the soil sample layering detection box 10 at a constant speed, and waits until the small animals are calm and stable. The in-situ soil living small animal layering example segmentation device divides the soil sample into k layers in the longitudinal position; the current soil sample layering serial number is q, and the serial number of the first layer of the top layer is q=1; for the first layer of soil sample, the information processing unit 63 in the in-situ soil living small animal layering example segmentation device controls the hyperspectral imaging module 7 combined with the broadband light source illumination module 6 to collect the reflectance spectrum of the small animals in the soil sample; the information processing unit 63 controls the hyperspectral imaging module 7 combined with the ultraviolet excitation module 20 to detect the fluorescence spectrum of the small animals in the soil sample.

[0046] The reflectance spectrum and the fluorescence spectrum constitute multi-modal hyperspectral imaging data, the soil animal 3D convolution instance segmentation deep network model is used to identify the species and the number of instance individuals of the small animals on the surface of the first layer of soil sample, and the number set T of species of the small animals on the surface of the current soil sample is obtained. 1 (T 1 1,T 1 2,T 1 3, …,T 1 m ); the small animals inside the soil sample are manually separated, selected and manually identified, and the total number set P of species of the small animals in the first layer of soil sample in the soil sample layering detection box 10 is obtained. 1 (P 1 1,P 1 2,P 1 3, …,P 1 m ), the machine learning P 1 (P 1 1,P 1 2,P 1 3, …,P 1 m) and T 1 (T 1 1,T 1 2,T 1 3, …,T 1 m ) of the fitting relationship.

[0047] The in-situ soil living small animal layer-by-layer instance segmentation device adopts layer-by-layer detection; after detecting the first layer of soil samples in the soil sample layer-by-layer detection box 10, the electric push arm module 8 pushes the soil samples in the soil sample layer-by-layer detection box into the soil sample storage box 11; the in-situ soil living small animal layer-by-layer instance segmentation device collects the multi-modal hyperspectral imaging data of the next layer of q-th layer of soil samples in the soil sample layer-by-layer detection box 10, identifies the species and instance individual number of small animals on the surface of the q-th layer of soil samples through the soil animal 3D convolution instance segmentation deep network model, and obtains the species number set T q (T q 1,T q 2,T q 3, …,T q m ) of the fitting relationship, and calculates the total small animal species number set P q (P q 1,P q 2,P q 3, …,P q m ) of the q-th layer according to the fitting relationship established above, wherein m is the total number of small animal species in the q-th layer of soil samples.

[0048] The above steps are sequentially repeated until all the layered soil samples in the soil sample layer-by-layer detection box 10 are detected. Finally, the species, distribution density and instance individual number of small animals in the in-situ soil sample and the longitudinal distribution of small animals in the soil are calculated.

[0049] The in-situ soil living small animal layer-by-layer instance segmentation device is shown in FIG. Figure 2

[0050] The in-situ soil living small animal layer-by-layer instance segmentation device comprises an instrument light-shielding airtight cover 1, a broadband light source illumination module 6, a hyperspectral imaging module 7, a first electric displacement table 12, an electric push arm module 8, a gas anesthesia system 9, a soil sample layer-by-layer detection box 10, a soil sample storage box 11, an ultraviolet light excitation module 20, an information processing unit 63, and a flipable transparent window module 64; the broadband light source illumination module 6, the hyperspectral imaging module 7, the first electric displacement table 12, the electric push arm module 8, the gas anesthesia system 9, and the information processing unit are connected through data lines.

[0051] ​The wideband light source illumination module 6 includes a first wide spectrum light source 2, a second wide spectrum light source 3, a first condenser lens 4, and a second condenser lens 5; the wideband light source illumination module 6 provides wideband illumination for the soil sample of the soil sample layered detection box 10, so that the hyperspectral imaging module 7 collects the reflected hyperspectral data of the soil sample.

[0052] The hyperspectral imaging module 7 includes a first lens 13, a slit 14, a second lens 15, a spectral conversion module 16, a third lens 17, a camera 18, and a second motorized displacement platform 19; the first lens 13, the slit 14, the second lens 15, the spectral conversion module 16, the third lens 17, the camera 18, and the second motorized displacement platform 19 are connected; the first motorized displacement platform 12 drives the movement of the hyperspectral imaging module.

[0053] The motorized push arm module 8 includes a first motorized push arm 49, a second motorized push arm 50, a third motorized push arm 51, a fourth motorized push arm 52, a fifth motorized push arm 53, a sixth motorized push arm 54, and a seventh motorized push arm 55; the first motorized push arm 49, the second motorized push arm 50, the third motorized push arm 51, the fourth motorized push arm 52, the fifth motorized push arm 53, the sixth motorized push arm 54, and the seventh motorized push arm 55 are sequentially connected from top to bottom.

[0054] The ultraviolet light excitation module 20 includes a first ultraviolet light source 21, a fourth lens 22, a first rotary motor 23, a first band-pass filter set 24, a second ultraviolet light source 25, a fifth lens 26, a second rotary motor 27, a second band-pass filter set 28, a third ultraviolet light source 29, a sixth lens 30, a third rotary motor 31, a third band-pass filter set 32, a fourth ultraviolet light source 33, a seventh lens 34, a fourth rotary motor 35, a fourth band-pass filter set 36, a fifth ultraviolet light source 37, an eighth lens 38, a fifth rotary motor 39, a fifth band-pass filter set 40, a sixth ultraviolet light source 41, a ninth lens 42, a sixth rotary motor 43, a sixth band-pass filter set 44, a seventh ultraviolet light source 45, a tenth lens 46, a seventh rotary motor 47, and a seventh band-pass filter set 48; the first ultraviolet light source 21, the fourth lens 22, and the first rotary motor 23 are connected to the first band-pass filter set 24; the second ultraviolet light source 25, the fifth lens 26, and the second rotary motor 27 are connected to the second band-pass filter set 28; the third ultraviolet light source 29, the sixth lens 30, and the third rotary motor 31 are connected to the third band-pass filter set 32; the fourth ultraviolet light source 33, the seventh lens 34, and the fourth rotary motor 35 are connected to the fourth band-pass filter set 36; the fifth ultraviolet light source 37, the eighth lens 38, and the fifth rotary motor 39 are connected to the fifth band-pass filter set 40; the sixth ultraviolet light source 41, the ninth lens 42, and the sixth rotary motor 43 are connected to the sixth band-pass filter set 44; and the seventh ultraviolet light source 45, the tenth lens 46, and the seventh rotary motor 47 are connected to the seventh band-pass filter set 48.

[0055] The flipable transparent window module 64 includes a first flipable transparent window 56, a second flipable transparent window 57, a third flipable transparent window 58, a fourth flipable transparent window 59, a fifth flipable transparent window 60, a sixth flipable transparent window 61, and a seventh flipable transparent window 62; the first flipable transparent window 56, the second flipable transparent window 57, the third flipable transparent window 58, the fourth flipable transparent window 59, the fifth flipable transparent window 60, the sixth flipable transparent window 61, and the seventh flipable transparent window 62 are connected from top to bottom.

[0056] The information processing unit 63 includes an industrial computer and a data connection line; the information processing unit 63 is connected with the broadband light source illumination module 6, the hyperspectral imaging module 7, the electric push arm module 8, the gas anesthesia system 9, and the ultraviolet light excitation module 20. The information processing unit 63 is used to control the operation timing of the broadband light source illumination module 6, the ultraviolet light excitation module 20, the hyperspectral imaging module 7, the electric push arm module 8, the gas anesthesia system 9, and the first electric displacement table 12, so that the hyperspectral imaging module 7 can orderly collect the multi-modal hyperspectral data of all the layered soil samples in the soil sample layered detection box 10. And the multi-modal hyperspectral data of the soil sample is used for instance segmentation of small animals to calculate the information such as the species, the number of instance individuals, and the distribution density of the small animals in the soil sample.

[0057] The layered detection method, the soil animal 3D convolution instance segmentation deep network model calculates the species and the number of instance individuals of the small animals on the soil surface according to the multi-modal hyperspectral imaging data composed of the reflectance spectrum and the fluorescence spectrum collected by the in-situ soil living small animal layered instance segmentation device, and estimates the species and the number of instance individuals of the small animals in the deep layer of the soil by using the following method.

[0058] The volume of the soil sample obtained from the soil in-situ is (x, y, z), and the number of small animals of various species in the soil is S (S1, S2, S 3, …, S n , where n is the number of species of small animals in the soil.

[0059] The in-situ soil living small animal layered instance segmentation device divides the soil sample into k layers in the longitudinal position; it is assumed that the current soil sample layered serial number is q, the volume of each layer of soil sample is (x, y, z / k), and the number of small animals of various species in the qth layer of soil sample is P q (P q 1, P q 2, P q 3, …, P q m), where m is the total number of small animal species in the qth layer of soil samples.

[0060] The calibration method is as follows: based on the reflectance spectrum and fluorescence spectrum of the surface of the first layer of soil samples in the soil sample stratification detection box 10 collected by the in-situ soil living small animal stratification instance segmentation device, the multimodal hyperspectral imaging data is formed, and the species and instance number of small animals on the surface of the first layer of soil samples are identified through the soil animal 3D convolution instance segmentation deep network model, and the species and number of small animals on the surface of the current soil sample are obtained. 1 (T 1 1,T 1 2,T 1 3, …,T 1 m By manually separating the small animals inside the soil samples, selecting and manually identifying, the total number of small animal species in the first layer of soil samples 10 in the soil sample detection box stratified set P 1 (P 1 1,P 1 2,P 1 3, …,P 1 m ); Through machine learning P 1 (P 1 1,P 1 2,P 1 3, …,P 1 m ) and T 1 (T 1 1,T 1 2,T 1 3, …,T 1 m ) is as follows:

[0061]

[0062] According to the above fitting relationship, the dataset T of small animals on the surface of the qth layer of soil samples is obtained by combining the in-situ soil living small animal layer instance segmentation device with the soil animal 3D convolutional instance segmentation deep network model. q (T q 1,T q 2,T q 3, …,T q m ), the total small animal species dataset P of the qth layer soil sample can be calculated q (P q 1,P q 2,Pq 3, …,P q m )。

[0063] The distribution density of the soil small animals of the i-th species in the q-th layer of the soil sample in the soil sample layered detection box 10 is:

[0064]

[0065] The distribution density of the soil small animals in the q-th layer of the soil sample in the soil sample layered detection box 10 is:

[0066]

[0067] According to the above steps, the data set of the species of the small animals in the soil sample in the soil sample layered detection box 10 is sequentially cycled until all the data sets of the species of the small animals in the soil sample in the soil sample layered detection box 10 are detected.

[0068] The number of the individuals of the soil small animals of various species in the soil sample layered detection box 10 is:

[0069]

[0070] The total number of the individuals of the soil small animals in the soil sample layered detection box 10 is:

[0071]

[0072] The distribution density of the soil small animals of various species in the soil sample layered detection box 10 is:

[0073]

[0074] The total distribution density of the soil small animals is:

[0075]

[0076] The soil animal 3D convolution instance segmentation deep network model can extract the spatial and spectral features of the multi-modal hyperspectral imaging data collected by the in-situ soil living small animal layered instance segmentation device based on the 3D convolution neural network, so as to perform instance segmentation on the small animals in the soil. The instance segmentation species individual information can be used to analyze the growth cycle, individual information such as male and female of the soil small animal individuals, so as to better establish the soil small animal library. Even if the small animals are inside the soil, due to the shielding of the soil, the actual collected soil small animal individuals are incomplete, and through the soil animal 3D convolution instance segmentation deep network model defined by the present application, the incomplete soil small animals can be instance segmented, which can not only segment and identify the species and instance individuals of the small animals, but also analyze whether the incomplete small animal parts are covered living small animals or amputated dead bodies.

[0077] The layered detection, the hyperspectral imaging module 7 in the in-situ soil living small animal layered example segmentation device detects the example segmentation of the individual of the first layer of soil small animals, the first electric push arm 49 in the electric push arm module 8 pushes the first layer of soil through the first movable transparent window into the soil sample storage box 11; the first electric displacement table 12 in the hyperspectral imaging module 7 drives the hyperspectral imaging module 7 to move downwards to make the second layer of soil surface imaging clear; the hyperspectral imaging module 7 detects the example segmentation of the individual of the second layer of soil small animals, the second electric push arm 50 in the electric push arm module 8 pushes the second layer of soil through the second movable transparent window into the soil sample storage box 11; the cycle is repeated in turn until all the layered soil samples in the soil sample layered detection box 10 are detected.

[0078] The multi-modal hyperspectral imaging detection, the broadband light source illumination module 6 adopts a wide spectral band of visible-near infrared band 400-1800nm to detect soil small animals, after detecting the reflectance spectrum of the soil sample. The information processing center controls to turn off the power supply of the broadband light source illumination module 6 and turn on the power supply of the ultraviolet excitation module 20; when the first layer of soil sample in the soil sample layered detection box 10 is detected, the ultraviolet light emitted by the first ultraviolet light source 21 is collimated through the fourth lens 22 and then passes through the first band-pass filter set 24 to irradiate the surface of the soil small animals from the side, and the soil small animals emit fluorescence after being excited by the ultraviolet light, and the fluorescence spectrum data is collected by the hyperspectral imaging module 7; the first rotating motor 23 drives the first band-pass filter set 24 to rotate to switch different band-pass filters and realize different fluorescence wavelength bands; when the next layer of soil sample in the soil sample layered detection box 10 is detected, the broadband light source illumination process and the switching of the second ultraviolet light source 25 are repeated to make the ultraviolet light irradiate the next layer of soil sample; the cycle is repeated in turn until the multi-modal hyperspectral imaging detection of all the layered soil samples in the soil sample layered detection box 10 is completed.

[0079] The reflectance or fluorescence image of the surface layer soil sample in the soil sample layered detection box 10 of the hyperspectral imaging module 7 passes through the first lens 13 and is imaged on the slit surface, and the light passing through the center of the slit is collimated into parallel light of different angles after passing through the second lens 15; the parallel light of different angles passes through the third lens 17 after passing through the spectral transformation module, and is focused on the camera 18 to form a spectral image; the spectral transformation module is composed of a prism-grating-prism module or a tunable filter, which is used to divide the complex light into monochromatic light of different wave bands; the second electric displacement platform 19 drives the hyperspectral imaging module 7 to perform one-dimensional scanning, thereby realizing the reflectance-fluorescence multi-modal hyperspectral imaging detection of the surface layer soil sample in the soil sample layered detection box.

[0080] Application examples

[0081] Figure 3 Fig. 7 is an example segmentation effect diagram of soil small animals by the in-situ soil living small animal hierarchical instance segmentation device. Soil small animals in the in-situ soil sample, including sow bugs 70, sow bugs 71, crustaceans 72, whip scorpions 73, geophilic centipedes 74, stone centipedes 75, millipedes 76, and crickets 77, are recognized and classified.

[0082] Through the in-situ soil living small animal hierarchical instance segmentation device, multi-modal hyperspectral data of in-situ soil small animals is collected. From the single-band reflection image, due to the complex scene covered by soil, small animals in the soil are mixed in the soil, and only a small part is exposed on the soil surface. It is extremely difficult to perform pixel region segmentation on small animals only by the small part of the worm body exposed on the soil surface. The soil animal 3D convolution instance segmentation deep network model proposed in the present application can accurately perform pixel region segmentation on small animals in the soil based on the collected multi-modal hyperspectral data, directly identify the small animal species, density, and longitudinal spatial distribution of each pixel, and provide more reliable data support for soil ecological research and agricultural management. It provides an advanced tool for soil ecological research and agricultural practice to promote soil health management and sustainable agricultural development.

[0083] The embodiments in the above description can be further combined or replaced, and the embodiments are only used to describe the preferred embodiments of the present application, and do not limit the concept and scope of the present application. Without departing from the design idea of the present application, various changes and improvements to the technical solutions of the present application made by those skilled in the art are within the protection scope of the present application. The protection scope of the present application is given by the appended claims and any equivalents thereof.

Claims

1. A method for in-situ soil micro-animal layered detection and investigation, characterized by: A naturally collected soil sample is placed in an in-situ soil live small animal stratification instance segmentation device, and a gas anesthesia system controls the anesthetic gas to flow into the soil sample stratification detection box (10) at a uniform speed. After the small animals in the soil sample are anesthetized, the information processing unit (63) in the in-situ soil live small animal stratification instance segmentation device controls the hyperspectral imaging module (7) in combination with the broadband light source illumination module (6) to collect the reflection spectrum of the small animals in the soil sample; the information processing unit (63) controls the hyperspectral imaging module (7) in combination with the ultraviolet light excitation module (20) to detect the fluorescence spectrum of the small animals in the soil sample; the reflection spectrum and the fluorescence spectrum constitute multimodal hyperspectral imaging data, and the soil animal 3D convolution instance segmentation deep network model performs pixel instance segmentation on the data, and simultaneously segments the soil small animals and the pixel area of ​​the soil, identifies the pixel label category of the soil small animals; The in-situ soil animal stratification instance segmentation device uses stratified detection, and the soil animal 3D convolutional instance segmentation deep network model independently identifies the species and number of small animals in each layer of soil. Finally, the species, number of small animals, distribution density, and vertical distribution of small animals in the entire in-situ soil sample are calculated. The in-situ soil living small animal stratification instance segmentation device comprises an instrument light-shielding sealed cover (1), a broadband light source illumination module (6), a hyperspectral imaging module (7), a first electric displacement stage (12), an electric push arm module (8), a gas anesthesia system (9), a soil sample stratification detection box (10), a soil sample storage box (11), an ultraviolet light excitation module (20), an information processing unit (63), and a flippable transparent window module (64); the broadband light source illumination module (6), the hyperspectral imaging module (7), the first electric displacement stage (12), the electric push arm module (8), and the gas anesthesia system (9) are respectively connected to the information processing unit.

2. The layered detection and investigation method according to claim 1, characterized in that: The soil animal 3D convolutional instance segmentation deep network model calculates the species and instance number of small animals in the soil surface layer based on the reflectance and fluorescence spectra collected by the in-situ soil living small animal layered instance segmentation device, and estimates the species and instance number of small animals in the deep soil layer using the following method: The volume of the soil sample obtained from the soil in situ is (x, y, z), and the number of various small animals in the soil is S (S1, S2, S 3, …, S n ), where n is the number of small animal species in the soil; The in-situ soil living small animal stratification segmentation device divides the soil sample into k layers in the longitudinal position; let the current soil sample layer number be q, the volume of each layer of soil sample be (x, y, z / k), and the number of various small animals in the qth layer of soil sample be P q (P q 1, P q 2, P q 3, …, P q m ), where m is the total number of small animal species in the qth layer of soil sample; The calibration method is as follows: based on the reflection spectrum and fluorescence spectrum of the surface of the first layer of soil sample in the soil sample stratification detection box (10) collected by the in-situ soil living small animal stratification instance segmentation device, the multimodal hyperspectral imaging data is formed, and the species and instance individual number of the small animals on the surface of the first layer of soil sample are identified through the soil animal 3D convolution instance segmentation deep network model, and the species and number of small animals on the surface of the current soil sample are obtained. 1 (T 1 1, T 1 2, T 1 3, …, T 1 m ); By manually separating the small animals inside the soil sample, selecting and manually identifying them, the total number of small animal species in the first layer of the soil sample in the soil sample stratification detection box (10) is obtained. 1 (P 1 1, P 1 2, P 1 3, …, P 1 m ); Through machine learning P 1 (P 1 1,P 1 2, P 1 3, …, P 1 m ) and T 1 (T 1 1, T 1 2, T 1 3, …, T 1 m ) is as follows: ； According to the above fitting relationship, the dataset T of small animals on the surface of the qth layer of soil samples is obtained by combining the in-situ soil living small animal layer instance segmentation device with the soil animal 3D convolutional instance segmentation deep network model. q (T q 1, T q 2,T q 3, …, T q m ), the total small animal species dataset P of the qth layer soil sample can be calculated q (P q 1, P q 2, P q 3, …, P q m ); Then, in the soil sample stratification detection box (10), the distribution density of soil small animals of species i in the qth layer of soil sample is: ； Then, in the soil sample layer detection box (10), the distribution density of soil small animals in the qth layer of soil sample is: ； The above steps are repeated in sequence until all the small animal species data sets of the stratified soil samples in the soil sample stratification detection box (10) are detected; The number of individuals of various types of soil microfauna in the soil sample stratification detection box (10) is: ； The total number of soil microfauna in the soil sample stratification test box (10) is: ； The distribution density of various soil microfauna species in the soil sample stratification detection box (10) is: ； The total distribution density of soil microfauna is: 。 3. The layered detection and investigation method according to claim 1, characterized in that: The soil animal 3D convolutional instance segmentation deep network model is based on a 3D convolutional neural network and extracts the spatial and spectral features of multimodal hyperspectral imaging data collected by an in-situ stratified instance segmentation device for living soil animals to perform instance segmentation of soil animals. The individual pixel information of the instance segmented species is used to analyze the growth cycle and gender information of the individual soil animals.

4. The layered detection and investigation method according to claim 1, characterized in that: When the hyperspectral imaging module (7) in the in-situ soil living small animal stratification instance segmentation device detects the instance segmented species of the first layer of soil small animals, the first electric push arm (49) in the electric push arm module (8) pushes the first layer of soil through the first flip-up transparent window into the soil sample storage box (11); the first electric displacement platform (12) in the hyperspectral imaging module (7) drives the hyperspectral imaging module (7) to move downward so that the surface image of the second layer of soil is clear; when the hyperspectral imaging module (7) detects the instance segmented species of the second layer of soil small animals, the second electric push arm (50) in the electric push arm module (8) pushes the second layer of soil through the second flip-up transparent window into the soil sample storage box (11); the cycle is repeated until all the stratified soil samples in the soil sample stratification detection box (10) are detected.

5. The layered detection and investigation method according to claim 1, characterized in that: In the multimodal hyperspectral imaging detection, the broadband light source illumination module (6) uses a wide spectrum band visible-near infrared band of 400-1800 nm to detect soil small animals. After detecting the reflection spectrum of the soil sample, the information processing center controls to turn off the power of the broadband light source illumination module (6) and turn on the power of the ultraviolet light excitation module (20). When detecting the first layer of soil samples in the soil sample stratification detection box (10), the ultraviolet light emitted by the first ultraviolet light source (21) is collimated by the fourth lens (22) and passes through the first bandpass filter group (24) and then irradiated from the side to the surface of the soil small animals. The soil small animals are excited by the ultraviolet light to produce fluorescence and are then detected by the hyperspectral imaging module. (7) collecting fluorescence spectrum data; the first rotating motor (23) drives the first bandpass filter group (24) to rotate, so as to increase the switching of different bandpass filters and realize the emission of different fluorescence bands; when detecting the next layer of soil samples in the soil sample stratification detection box (10), repeating the broadband light source illumination process and switching the second ultraviolet light source (25) to allow the ultraviolet light to irradiate the next layer of soil samples; the cycle is repeated until the multimodal hyperspectral imaging detection of all stratified soil samples in the soil sample stratification detection box (10) is completed.

6. The layered detection and investigation method according to claim 1, characterized in that: The hyperspectral imaging module (7) is characterized in that the reflection or fluorescence image of the surface soil sample in the soil sample stratification detection box (10) is imaged on the slit surface after passing through the first lens (13), and the light passing through the center of the slit is collimated into parallel light of different angles after passing through the second lens (15); the parallel light of different angles passes through the spectrum conversion module and then through the third lens (17) to focus on the camera (18) to form a spectrum image; the spectrum conversion module is composed of a prism-grating-prism module, or a tunable filter, and is used to separate the complex light into monochromatic light of different bands; the second electric displacement platform (19) drives the hyperspectral imaging module (7) to perform one-dimensional scanning, thereby realizing the reflection-fluorescence multimodal hyperspectral imaging detection of the surface soil sample in the soil sample stratification detection box.

7. An in-situ soil living small animal layer instance segmentation device, characterized by: Applied to the layered detection and investigation method according to claim 1; The broadband light source lighting module (6) comprises a first broadband light source (2), a second broadband light source (3), a first condensing lens (4), and a second condensing lens (5); the first broadband light source (2) is connected to the second condensing lens (5); the second broadband light source (3) is connected to the first condensing lens; The hyperspectral imaging module (7) includes a first lens (13), a slit (14), a second lens (15), a spectrum conversion module (16), a third lens (17), a camera (18), and a second electric displacement platform (19); the first lens (13), the slit (14), the second lens (15), the spectrum conversion module (16), the third lens (17), and the camera (18) are connected to the second electric displacement platform (19); and the first electric displacement platform (12) drives the hyperspectral imaging module to move; The electric push arm module (8) includes a first electric push arm (49), a second electric push arm (50), a third electric push arm (51), a fourth electric push arm (52), a fifth electric push arm (53), a sixth electric push arm (54), and a seventh electric push arm (55); the first electric push arm (49), the second electric push arm (50), the third electric push arm (51), the fourth electric push arm (52), the fifth electric push arm (53), the sixth electric push arm (54), and the seventh electric push arm (55) are connected in sequence from top to bottom; The ultraviolet light excitation module (20) comprises a first ultraviolet light source (21), a fourth lens (22), a first rotating motor (23), a first bandpass filter group (24), a second ultraviolet light source (25), a fifth lens (26), a second rotating motor (27), a second bandpass filter group (28), a third ultraviolet light source (29), a sixth lens (30), a third rotating motor (31), a third bandpass filter group (32), a fourth ultraviolet light source (33), a seventh lens (34), a fourth rotating motor (35), a fourth bandpass filter group (36), a fifth ultraviolet light source (37), an eighth lens (38), a fifth rotating motor (39), a fifth bandpass filter group (40), a sixth ultraviolet light source (41), a ninth lens (42), a sixth rotating motor (43), a sixth bandpass filter group (44), a seventh ultraviolet light source (45), a tenth lens (46), a seventh rotating motor (47), and a seventh bandpass filter group (48). The first ultraviolet light source (21), the fourth lens (22), the first rotating motor (23) are connected to the first bandpass filter group (24); the second ultraviolet light source (25), the fifth lens (26), the second rotating motor (27) are connected to the second bandpass filter group (28); the third ultraviolet light source (29), the sixth lens (30), the third rotating motor (31) are connected to the third bandpass filter group (32); the fourth ultraviolet light source (33), the seventh lens (34), the fourth The rotating motor (35) is connected to the fourth band-pass filter group (36); the fifth ultraviolet light source (37), the eighth lens (38), and the fifth rotating motor (39) are connected to the fifth band-pass filter group (40); the sixth ultraviolet light source (41), the ninth lens (42), and the sixth rotating motor (43) are connected to the sixth band-pass filter group (44); and the seventh ultraviolet light source (45), the tenth lens (46), and the seventh rotating motor (47) are connected to the seventh band-pass filter group (48); The flip-able transparent window module (64) comprises a first flip-able transparent window (56), a second flip-able transparent window (57), a third flip-able transparent window (58), a fourth flip-able transparent window (59), a fifth flip-able transparent window (60), a sixth flip-able transparent window (61), and a seventh flip-able transparent window (62); the first flip-able transparent window (56), the second flip-able transparent window (57), the third flip-able transparent window (58), the fourth flip-able transparent window (59), the fifth flip-able transparent window (60), the sixth flip-able transparent window (61), and the seventh flip-able transparent window (62) are connected from top to bottom; The information processing unit (63) includes an industrial computer and a data connection line; the information processing unit is respectively connected to the broadband light source illumination module (6), the hyperspectral imaging module (7), the electric push arm module (8), the gas anesthesia system (9), the ultraviolet light excitation module (20) and the first electric translation stage (12).

Images