CN-121982417-A - Portable cordyceps sinensis detection method

CN121982417ACN 121982417 ACN121982417 ACN 121982417ACN-121982417-A

Abstract

The invention discloses a portable cordyceps sinensis detection method, which relates to the technical field of image recognition and comprises the steps of collecting a scene image of a suspected cordyceps sinensis region through a high-definition camera of a portable collecting terminal, preprocessing and compressing the scene image, transmitting the coded image to an edge computing terminal through wireless communication, decoding the coded image to obtain an image to be detected, detecting a cordyceps sinensis target, outputting a detection result comprising candidate target positions, types and confidence degrees, transmitting the detection result back to the collecting terminal for display, generating a collecting parameter adjustment instruction based on the detection result by the edge computing terminal, transmitting the collecting parameter adjustment instruction to the collecting terminal, performing feedback control on at least one of exposure, focusing and collecting frequency, caching the image and the detection result, and uploading a cloud end to establish an index when uploading conditions are met. The portable rapid detection of the end-edge coordination is realized, the gate control is carried out through the preprocessing configuration mark, the closed loop misadjustment parameter caused by the preprocessing switching is restrained, and the detection stability and the traceability are improved.

Inventors

HU JUNHUA
JIN WEIZHENG
LI MUYUN

Assignees

深圳市智坤动力科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260204

Claims (10)

1. A portable cordyceps sinensis detection method, comprising: acquiring a scene image of a suspected cordyceps area through a high-definition camera of a portable acquisition terminal; preprocessing and compression encoding are carried out on the scene image at the acquisition terminal; Transmitting the coded image to an edge computing end through wireless communication; Decoding the coded image at the edge computing end to obtain an image to be detected, detecting a cordyceps sinensis target of the image to be detected, outputting a detection result and transmitting the detection result back to the acquisition terminal for display, wherein the detection result comprises the position, the category and the confidence information of the candidate target; The edge computing end generates an acquisition parameter adjustment instruction based on the detection result, transmits the acquisition parameter adjustment instruction to the acquisition terminal through wireless communication, and performs feedback control on at least one of exposure, focusing and acquisition frequency of the acquisition terminal; And (3) caching the image and the detection result in a local storage of the acquisition terminal, uploading the cloud end when the uploading condition is met, and establishing an index according to the acquisition time, the acquisition place or the user number.
2. The method according to claim 1, wherein preprocessing the scene image includes performing at least two of scaling, cropping, color space conversion, and distortion correction on the scene image, and wherein the preprocessed image is used as an input for compression encoding.
3. The method according to claim 1, wherein the compression encoding includes encoding the preprocessed image to generate encoded image data, and writing frame identification information for feedback association into the encoded image data, so that the edge computing end establishes a correspondence between the detection result and the corresponding image.
4. The portable worm grass detection method according to claim 1, wherein the position of the worm grass target detected at the edge calculating end includes at least the boundary frame coordinates or outline area information of the candidate target, and the confidence information includes a confidence value or confidence level for characterizing the confidence level that the candidate target is worm grass.
5. The method according to claim 1, wherein the feedback control includes determining an acquisition parameter adjustment command of the acquisition terminal based on the confidence information and a position stability of the candidate target, the acquisition parameter adjustment command including at least one of an exposure adjustment command, a focus adjustment command, and an acquisition frequency adjustment command, wherein the position stability is obtained by performing a logic comparison based on position information of the same candidate target in adjacent acquisition rounds, and the logic comparison includes at least any one of intersection, inclusion, or proximity of bounding boxes of the candidate target.
6. The method according to claim 5, wherein the exposure adjustment command includes at least one of either increasing exposure or decreasing exposure, the focus adjustment command includes at least one of refocusing or adjusting focus position, and the collection frequency adjustment command includes at least one of either increasing collection frequency or decreasing collection frequency.
7. The method according to claim 1, wherein the preprocessing configuration mark is generated when the acquisition terminal preprocesses and compression-encodes the scene image, and the preprocessing configuration mark is used for representing a link enabling state and a preset configuration of preprocessing, wherein the link enabling state and the preset configuration at least comprise at least one of an enabling state and a preset configuration of geometric normalization processing, an enabling state and a preset configuration of color gamut normalization processing, and an enabling state and a preset configuration of compression-encoding processing.
8. The method according to claim 7, wherein the image of the preprocessed configuration mark after encoding is transmitted to an edge computing end through wireless communication, the edge computing end binds the preprocessed configuration mark with the detection result when the detection result is generated, and the preprocessed configuration mark is transmitted back to the acquisition terminal for displaying the detection result of different acquisition rounds in a correlated manner.
9. The method according to claim 7, wherein the edge computing terminal determines a preprocessing isomorphic state based on a logical comparison of preprocessing configuration markers of adjacent collection rounds, the preprocessing isomorphic state at least comprising an isomorphic state, a transitional state and a heterogeneous state; determining an isomorphic state when the preprocessing configuration marks of adjacent acquisition rounds are consistent; When the preprocessing configuration mark of the current collection round is switched relative to the previous collection round, the corresponding state of the current collection round is recorded as a transition state to be judged, if the preprocessing configuration mark of the next collection round is consistent with the current collection round, the transition state to be judged is confirmed as a transition state, and if the preprocessing configuration mark of the next collection round is switched again, the transition state to be judged is confirmed as a heterogeneous state, wherein the heterogeneous state is a state that the preprocessing configuration mark is switched at least twice in the continuous collection round.
10. The method according to claim 9, wherein the feedback control is performed based on the detection result when the isomorphic state is determined; Freezing at least one of an exposure adjustment instruction and a focus adjustment instruction based on the detection result when in the transitional state, and allowing only the acquisition frequency adjustment instruction to be executed based on the detection result or keeping the acquisition parameters unchanged; when the heterogeneous state is determined, the feedback control is forbidden according to the detection result, and the edge computing end generates a preprocessing configuration locking instruction and transmits the preprocessing configuration locking instruction to the acquisition terminal, so that the acquisition terminal keeps the preprocessing link preset configuration unchanged in at least one subsequent acquisition round; And writing the corresponding preprocessing configuration mark, preprocessing isomorphic state, image and detection result into a local storage or uploading to the cloud.

Description

Portable cordyceps sinensis detection method Technical Field The invention relates to the technical field of image recognition, in particular to a portable cordyceps sinensis detection method. Background Cordyceps sinensis is used as a common traditional Chinese medicine material, and has high medicinal value and economic value. In links such as field collection, artificial culture inspection, flow quality control and the like, the identification, screening and grading of the cordyceps sinensis are dependent on artificial visual inspection and experience discrimination for a long time. The method is easy to be influenced by experience difference and fatigue of operators under the conditions of small target size, fine form difference, complex background texture, strong camouflage property and the like, has the problems of low detection efficiency, strong subjectivity and poor result consistency, and is difficult to meet the requirements of real-time performance and traceability of large-scale acquisition and continuous quality monitoring. With the development of mobile terminals and image recognition technologies, image-based agricultural product detection schemes have been gradually applied to field scenes. However, the existing scheme mainly uses the general image acquisition equipment and the upper computer or the cloud server as centers, so that the system is large in size, high in deployment cost and high in dependence on a network environment. In the field or weak network environment, the image uploading, decoding and recognition calculation link is easy to have delay fluctuation or interruption, so that the detection process is difficult to stably and continuously execute, and partial mobile terminal application only provides a simple image classification or static recognition function and does not carry out terminal edge collaborative optimization around the small-size characteristic, background interference and on-site shooting uncertainty of the cordyceps sinensis target, so that detection omission and erroneous judgment still easily occur under the condition that complex illumination, gesture change are similar to the background. Further, in portable field detection, image acquisition quality is significantly affected by factors such as illumination, hand-held jitter, focus offset, and acquisition frequency settings. The existing scheme generally regards acquisition and identification as mutually independent stages, lacks a mechanism for performing closed-loop feedback adjustment on exposure, focusing and acquisition frequency of an acquisition end based on a detection result, so that when the detection confidence is reduced or the target position is unstable, the detection reliability is difficult to improve the detectability and stability of a subsequent frame in time through acquisition parameter adjustment. In addition, the field system often needs to perform preprocessing and transmission optimization such as scaling, clipping, color space conversion, distortion correction, compression coding and the like, but the existing scheme generally lacks explicit identification and management of a preprocessing link starting state and preset configuration, so that when the preprocessing configuration is switched in a continuous acquisition round, the fluctuation source of a detection result is difficult to distinguish, and further false triggering or oscillation of acquisition parameter feedback control can be caused, and the overall detection stability and engineering usability are affected. Disclosure of Invention Based on the above-mentioned shortcomings of the prior art, the present invention aims to provide a portable Cordyceps detection method, so as to solve the above-mentioned technical problems. In order to achieve the above purpose, the invention provides a portable cordyceps sinensis detection method, which comprises the following steps: acquiring a scene image of a suspected cordyceps area through a high-definition camera of a portable acquisition terminal; preprocessing and compression encoding are carried out on the scene image at the acquisition terminal; Transmitting the coded image to an edge computing end through wireless communication; Decoding the coded image at the edge computing end to obtain an image to be detected, detecting a cordyceps sinensis target of the image to be detected, outputting a detection result and transmitting the detection result back to the acquisition terminal for display, wherein the detection result comprises the position, the category and the confidence information of the candidate target; The edge computing end generates an acquisition parameter adjustment instruction based on the detection result, transmits the acquisition parameter adjustment instruction to the acquisition terminal through wireless communication, and performs feedback control on at least one of exposure, focusing and acquisition frequency of the acquisition terminal; And (3) caching th