CN-121982513-A - Red tide outbreak prediction method and system

CN121982513ACN 121982513 ACN121982513 ACN 121982513ACN-121982513-A

Abstract

The invention provides a red tide outbreak prediction method and a system, which relate to the technical field of red tide prediction and comprise the steps of firstly acquiring multi-space-time multi-source remote sensing data and corresponding real labels thereof; the multi-time-space multi-source remote sensing data are preprocessed to obtain preprocessed multi-time-space multi-source remote sensing data, the preprocessed multi-time-space multi-source remote sensing data are input into a constructed red tide outbreak prediction network to obtain a prediction label, a total loss function is constructed according to the real label and the prediction label, the red tide outbreak prediction network is trained to obtain a trained red tide outbreak prediction network, and multi-source remote sensing data to be predicted are input into the trained red tide outbreak prediction network to obtain red tide outbreak prediction results. The method can effectively utilize the information of the characteristic dimension and the space dimension, realizes the accurate prediction of the red tide outbreak, and has high robustness.

Inventors

LIN XIAOBO
LI XINYAN
SUN SHAOJIE
WANG ZHEN
LUO CHENGYU
HUANG GUANXIAN
CHEN JINZHENG
ZHU JIANXING
ZHAO JUN
LIANG YUCHEN

Assignees

广东省国土资源测绘院
中山大学

Dates

Publication Date: 20260505
Application Date: 20251209

Claims (10)

1. A red tide outbreak prediction method, comprising: acquiring multi-space-time multi-source remote sensing data and corresponding real labels thereof; Preprocessing the multi-space-time multi-source remote sensing data to obtain preprocessed multi-space-time multi-source remote sensing data; inputting the preprocessed multi-space-time multi-source remote sensing data into a constructed red tide outbreak prediction network to obtain a prediction label; Constructing a total loss function according to the real label and the predictive label, and training the red tide outbreak prediction network to obtain a trained red tide outbreak prediction network; And inputting the multisource remote sensing data to be predicted into a trained red tide outbreak prediction network to obtain a red tide outbreak prediction result.
2. The red tide outbreak prediction method according to claim 1, wherein preprocessing the multi-space-time multi-source remote sensing data to obtain preprocessed multi-space-time multi-source remote sensing data comprises the following steps: performing time sequence alignment on the multi-time-space multi-source remote sensing data to obtain multi-time-space multi-source remote sensing data after time sequence alignment; performing space alignment on the time-sequence-aligned multi-space-time multi-source remote sensing data to obtain space-aligned multi-space-time multi-source remote sensing data; and converting the space-aligned multi-space-time multi-source remote sensing data into a preset standard format, and removing abnormal values and filling the data to obtain the preprocessed multi-space-time multi-source remote sensing data.
3. The red tide outbreak prediction method according to claim 1, wherein the red tide outbreak prediction network comprises a feature weighting layer, an initial convolution layer, a first encoder, a second encoder, a third encoder, a time fusion layer, a first upsampling layer, a first splicing layer, a first decoder, a second upsampling layer, a second splicing layer, a second decoder, a third upsampling layer, a third splicing layer and a final convolution layer which are connected in sequence; The output end of the initial convolution layer is also connected with the input end of the third splicing layer; the output end of the first encoder is also connected with the input end of the second splicing layer; The output end of the second encoder is also connected with the input end of the first splicing layer.
4. The method for predicting red tide outbreak according to claim 3, wherein the initial convolution layer comprises a preset number of mixed neural modules connected in sequence.
5. The red tide outbreak prediction method according to claim 4, wherein the first encoder, the second encoder and the third encoder have the same structure and comprise a maximum downsampling unit, a preset number of mixed nerve modules and CBAM modules which are connected in sequence; the hybrid neural module includes a 3 x 3 convolution, a batch normalization unit, a ReLU activation function, and a Dropout unit connected in sequence.
6. The red tide outbreak prediction method according to claim 5, wherein the CBAM module comprises a channel attention unit, a characteristic and channel weight product unit, a spatial attention unit, a characteristic and spatial weight product unit and a Clamp unit which are connected in sequence; The channel attention unit comprises a global average pooling subunit, a1×1 convolution, a ReLU activation function, a1×1 convolution and a Sigmoid activation function which are sequentially connected; the spatial attention unit comprises a channel mean value subunit, a channel maximum value subunit, a splicing subunit, a7 multiplied by 7 convolution and a Sigmoid activation function which are connected in sequence.
7. The red tide outbreak prediction method according to claim 3, wherein the first upsampling layer, the second upsampling layer and the third upsampling layer have the same structure and comprise an upsampling bilinear interpolation unit, a 3 x 3 convolution, a batch normalization unit, a ReLU activation function and a size adjustment unit which are sequentially connected.
8. A red tide outbreak prediction method according to claim 3, characterized in that the final convolution layer comprises a3 x 3 convolution, a batch normalization unit, a ReLU activation function, a Dropout unit and a1 x1 convolution connected in sequence.
9. The red tide outbreak prediction method according to claim 1, wherein the total loss function expression is as follows: Wherein, the 、、 Respectively is 、、 Linear combination weights of (a); = Wherein, the Represent the first The sample weights of the individual picture elements, Representation model pair number The probability that an individual picture element is predicted as a positive sample, Represent the first True labels of the individual pixels, the smooth represents a constant; Wherein, the Represent the first Focal loss terms for individual pixels; , represent the first The probability that a sample is identified by the model as its true class, , The modulation index representing the Focal is shown, Represent the first The binary cross entropy term of the individual picture elements, Wherein, the TP represents true positive, expressed as FP represents a false positive, expressed as FN represents false negative, and the expression is , And Representing the weight coefficient.
10. A red tide outbreak prediction system for implementing the red tide outbreak prediction method according to claims 1 to 9, characterized by comprising: The data acquisition module is used for acquiring multi-space-time multi-source remote sensing data and corresponding real labels thereof; The preprocessing module is used for preprocessing the multi-space-time multi-source remote sensing data to obtain preprocessed multi-space-time multi-source remote sensing data; The prediction label acquisition module is used for inputting the preprocessed multi-space-time multi-source remote sensing data into a constructed red tide outbreak prediction network to obtain a prediction label; The network training module is used for constructing a total loss function according to the real label and the predictive label, and training the red tide outbreak prediction network to obtain a trained red tide outbreak prediction network; and the red tide outbreak prediction module is used for inputting the multisource remote sensing data to be predicted into a trained red tide outbreak prediction network to obtain red tide outbreak prediction results.

Description

Red tide outbreak prediction method and system Technical Field The invention relates to the technical field of red tide prediction, in particular to a red tide outbreak prediction method and a red tide outbreak prediction system. Background Red tide refers to a harmful ecological phenomenon that some phytoplankton, protozoa or bacteria in the ocean proliferate or gather highly under specific environmental conditions, thereby causing discoloration of water, and the outbreak of the harmful ecological phenomenon can cause damage and harm to the marine ecosystem. At present, red tide prediction by using a deep learning model has become a mainstream research direction in the field. The general flow of the method is to acquire historical marine environment monitoring data or remote sensing image data, construct and train a deep neural network model, mine characteristic rules contained in the data, and further predict future red tide outbreak conditions by using a trained network. However, the existing network model is not used enough for information of characteristic dimension and space dimension in the data, so that red tide outbreak prediction is inaccurate. Disclosure of Invention The invention provides a red tide outbreak prediction method and a red tide outbreak prediction system, which aim to overcome the defect that the information of the characteristic dimension and the space dimension in data is not utilized enough and the red tide outbreak prediction part is accurate. In order to solve the technical problems, the technical scheme of the invention is as follows: the invention provides a red tide outbreak prediction method, which comprises the following steps: acquiring multi-space-time multi-source remote sensing data and corresponding real labels thereof; Preprocessing the multi-space-time multi-source remote sensing data to obtain preprocessed multi-space-time multi-source remote sensing data; inputting the preprocessed multi-space-time multi-source remote sensing data into a constructed red tide outbreak prediction network to obtain a prediction label; Constructing a total loss function according to the real label and the predictive label, and training the red tide outbreak prediction network to obtain a trained red tide outbreak prediction network; And inputting the multisource remote sensing data to be predicted into a trained red tide outbreak prediction network to obtain a red tide outbreak prediction result. Preferably, preprocessing the multi-space-time multi-source remote sensing data to obtain preprocessed multi-space-time multi-source remote sensing data, including: performing time sequence alignment on the multi-time-space multi-source remote sensing data to obtain multi-time-space multi-source remote sensing data after time sequence alignment; performing space alignment on the time-sequence-aligned multi-space-time multi-source remote sensing data to obtain space-aligned multi-space-time multi-source remote sensing data; and converting the space-aligned multi-space-time multi-source remote sensing data into a preset standard format, and removing abnormal values and filling the data to obtain the preprocessed multi-space-time multi-source remote sensing data. Preferably, the red tide outbreak prediction network comprises a characteristic weighting layer, an initial convolution layer, a first encoder, a second encoder, a third encoder, a time fusion layer, a first upsampling layer, a first splicing layer, a first decoder, a second upsampling layer, a second splicing layer, a second decoder, a third upsampling layer, a third splicing layer and a final convolution layer which are connected in sequence; The output end of the initial convolution layer is also connected with the input end of the third splicing layer; the output end of the first encoder is also connected with the input end of the second splicing layer; The output end of the second encoder is also connected with the input end of the first splicing layer. Preferably, the initial convolution layer comprises a preset number of hybrid neural modules connected in sequence. Preferably, the first encoder, the second encoder and the third encoder have the same structure and each comprise a maximum downsampling unit, a preset number of mixed neural modules and CBAM modules which are sequentially connected; the hybrid neural module includes a 3 x 3 convolution, a batch normalization unit, a ReLU activation function, and a Dropout unit connected in sequence. Preferably, the CBAM module includes a channel attention unit, a feature and channel weight product unit, a spatial attention unit, a feature and spatial weight product unit, and a Clamp unit, which are sequentially connected; the channel attention unit comprises a global average pooling subunit, a1×1 convolution, a ReLU activation function, a1×1 convolution and a Sigmoid activation function which are connected in sequence. The spatial attention unit comprises a channel mean value subunit,