Real-time detection system for black tea fermentation process
A technology for real-time detection and fermentation degree, applied in manufacturing computing systems, measuring devices, testing plant materials, etc., can solve the problems of discriminant efficiency, stability and accuracy that need to be verified, not fully explained, etc., to ensure real-time continuous image acquisition. , Solve the effect of interfering with image quality and achieve isolation
Patent Information
- Authority / Receiving Office
- CN · China
- Current Assignee / Owner
- Publication Date
- 2022-05-31
Smart Images

Figure 1
Abstract
Description
A real-time detection system for black tea fermentation process technical field The invention belongs to the black tea detection equipment technical field, be specifically related to a kind of black tea fermentation process real-time detection system. Background technique
[0002] The primary process of black tea processing is mainly withering, rolling, fermentation, and drying. Black tea fermentation is mainly based on polyphenols The enzymatic oxidation of tea is the main process of forming theaflavins, thearubigins and theabrownins, and the color of the tea leaves is green-yellow-red. It is the key to forming the quality characteristics of red leaves in black tea and black soup, and the degree of fermentation is insufficient or excessive, which seriously affects the quality of black tea. Currently, The degree of fermentation depends on human experience, or manual control of the process based on simple process parameters such as temperature, humidity and time. but ...
Examples
Embodiment Construction
[0075] (7) Build a PyTorch-based MobileNetV3 model.
[0076] Initial part: 1 convolutional layer, through 3x3 convolution, namely convolutional layer, BN layer, h-switch activation layer, in Large
[0077] The middle part: a plurality of convolutional layers, which are the network structures of a plurality of blocks (MobileBlock) containing a convolutional layer, different
[0078] The last part: the Squeeze operation is omitted, the Avg Pooling is advanced, and the 1×1 convolution is directly used instead of the full
[0081] (9) Encapsulate by software through the deep learning model and the traditional model established in step (8).
[0084] Furthermore, it should be understood that although this specification is described in terms of embodiments, not every embodiment includes only
[0085] The technology, shape and structural part not described in detail in the present invention are all known technology.