Research On Habitat Prediction Based On Multi-source Remote Sensing Data

At present,China’s marine fisheries industry is facing major challenges,which are mainly due to the sharp decline of the main economic fish resources in the offshore in recent years and the rising production costs of pelagic fisheries.At present,the global fishing competition is becoming more and more intense,and there is still a considerable gap between China and fishing powerhouses in terms of fishing equipment and technology,which has brought enormous survival pressure to both fishermen and fishing enterprises.In response to this situation,accurate fishery prediction can reduce the search time and fishery cost,while the distribution and quantity changes of marine fishery resources are closely related to the marine environment,the extensive coverage,long-term data collection,and almost real-time availability of satellite remote sensing information have become crucial in the preservation,management,and growth of marine fishery resources.However,there are still some drawbacks in the current research on the prediction of pelagic fishery,and most of the previous researches are still at the stage of using productlevel marine remote sensing data and serial network model with single-source data as input.With the widespread use of remote sensing technology,accessing raw remote sensing data has become relatively easy.However,the current research lacks the integration and practical application of multi-source fused data,which hinders progress in the field of fishery prediction.In addition,by reading a large amount of domestic and foreign related literature,it is found that most of the current research on fishery prediction is still stuck in using the current month environmental data to invert the fishery distribution of the current month,and lacks the prediction of the fishery distribution of future months.Considering the issues mentioned earlier,this paper aims to address the following research objectives.1.This paper describes the current development status of fishery forecasting at home and abroad,briefly describes t The basic theory of the three data processing algorithms and the basic composition and operation principle of the neural network are briefly described,and makes a brief introduction to the evaluation index CPUE of fishing efficiency and resource utilization and the objective evaluation index of fishery forecasting results.This paper presents a brief introduction of CPUE and objective evaluation indexes for fishing prediction,and prepares the ground for the subsequent research on fishing prediction.2.To address the problems of using a single environmental data source and the model is mostly a sequential network model with single-source data as input,this paper tries to use product-level remote sensing data and L1B-level raw remote sensing data to match the Northwest Pacific fallfish fishery dataset respectively to obtain multi-source heterogeneous experimental data.This paper proposes a fishery inversion approach that integrates multiple sources of diverse remote sensing data.The method comprises a feature extraction model based on a combination of convolutional neural network(CNN)and long short-term memory network(LSTM),as well as a decision-level fusion model that combines the extracted features from multiple sources of heterogeneous data.The experimental results demonstrate that the proposed model performs remarkably well,with an average R2 of 0.9901 and a low RMSE of 0.01588 in the model validation experiments.These results significantly surpass the outcomes obtained from any single-source model.In addition,the proposed method exhibits excellent generalization performance,keeping the inversion error of the test dataset below 8% in the generalization experiments,which demonstrates its robustness and good generalization ability.This study validates the applicability of remote sensing data for fishery inversion,and expands the range of data selection for fishery prediction research.It provides innovative insights for future research in this field.3.To address the problem of lack of prediction of fishery distribution in future months,this paper proposes to create a sample set with time series using product-level remote sensing data based on the existing experimental research,and to predict the fishery in current or future months by using temporally continuous environmental data of previous years or months,and innovatively constructs a fusion network model that can fully extract the sample set.The model is based on LSTM.The main framework of the model is based on the powerful ability of LSTM to extract continuous time series features,and the feature extraction is carried out separately followed by feature fusion according to the different data structure characteristics of each part of the constructed data,and more accurate model validation results are obtained.The experiment proves the sensitivity of fishery prediction to time series environmental data,verifies the feasibility of using time series environmental datasets for fishery prediction,and lays a technical foundation for the practical application of fishery prediction methods.