| The fisheries acoustic method is an important method in the field of biological resource detection because of its advantages of rapid and efficient,wide detection range,no harm to biological resources,and provision of continuous data.At present,the main platform for acoustic survey and monitoring of ocean living resources and related environmental factors is the research vessel equipped with quantitative scientific echosounder instrument.Although the use of research vessels for acoustic surveys can obtain accurate biological and environmental information for a certain period of time and a certain range of size,the cost of obtaining data is high,and it is difficult to reflect the changes of fish resources in large spatial-temporal scale.As the main carrier of marine fishery production,pelagic fishing vessels sail in various fishing grounds between oceans all year round.With the improvement of science and technology,most fishing vessels have been equipped with high-resolution commercial echo sounders.The acoustic data derived from them have already met the needs of scientific resource assessment.However,the exported commercial acoustic data cannot be directly applied to resource assessment and monitoring due to the problems of non-uniform original data format and calibration of data.The conversion of commercial acoustic data into valid scientific data is extremely critical.At the same time,with the expansion of ocean exploration range,using manual methods to extract bio-echo signals of interest from acoustic data is labor-intensive and the accuracy rate decreases with time.Deep learning algorithms,as an important method to assist manual processing of large amounts of data,can ensure accuracy while quickly identifying useful information.The application of depth learning algorithm to the recognition of bio-echo signal is becoming an important development direction of fishery acoustics research.This study is based on the acoustic data collected from commercial fishing vessel FV‘Ming Hua’in the Northwest Pacific Ocean from May 6 to August 12,2021.The automatic conversion method of commercial acoustic data and the pre-processing method of acoustic echogram in different underwater scenes are studied.A rapid calibration method for commercial echosounder is presented.The echo traces in the pre-processed acoustic data are filtered by depth learning algorithm YOLOv5,and the echo traces are classified according to the morphology of fish swarm.Based on the living fish calibration method,the commercial echosounder performance test using relative calibration method is accomplished,and the calibration values are obtained.Echo integration method was used to calculate the change of NASC value of pacific saury during the fishing process and the horizontal and vertical distribution in the survey area respectively.(1)According to the data format of commercial echo sounder,the program of automatic acoustic data conversion is written by MATLAB,and the original binary acoustic data is converted into readable acoustic data in CSV format,at the same time,the visualization of acoustic data is realized.In the post-processing of acoustic data,three different threshold setting methods are used to remove noise and reverberation in the echograms of different underwater scenes.The results show that the fixed threshold processing method and the adaptive threshold processing method can remove most of the noise and retain the target echo traces,while the OTSU threshold processing method cannot completely split the noise and the target echo traces.The fixed thresholding method is more effective for reverberant signals,while the adaptive thresholding method can preserve the morphological characteristics of the echo trace more completely.(2)Echo traces were extracted and classified based on the pre-processed images,and the YOLOv5 algorithm was used to train an automatic echo trace recognition model.Experimental results:A total of 10710 echo traces were extracted from the dataset,and the echo traces were classified as 1-single fish,2-multiple fish,and 3-fish schools,and segmented into training set(85%),validation set(5%),and test set(15%).The model obtained after training has a recognition accuracy of 0.79,a recognition recall of 0.68,and an F1 value of 0.66 obtained within a confidence interval of 0.454,which is shown that the model can accurately identify most of the echo traces.The difference coefficients of the transducer transmitting and receiving living fish acoustic signals were calculated using the identified extracted single fish echo traces and the prolate spheroid model(PSM),and compared with the standard sphere difference coefficients.The results showed that the difference coefficients obtained by the living fish calibration method and the standard sphere difference coefficients were 1.35 d B and 0.54 d B,which were within the normal range,and the acoustic data collected by the echo sounder could be applied to fish stock assessment.The experiment also proves the feasibility of rapid calibration of commercial echosounder.(3)The spatial distribution of pacific saury resources in the fishing area was analyzed using acoustic data during the June and July fishing period in 2021.Based on the acoustic data during the June and July fishing periods,we found that the average fishing time was 24 min in June and 20 min in July.the range of NASC values during the June fishing period was 1662～3742 m~2/nmi~2and the range of NASC values during the July fishing period was 7032～13022 m~2/nmi~2,while the catches in June were more than those in July.This suggests the possibility of a larger number of mixed species with similar target intensities in the acoustic data,as well as the presence of a large number of age-0 fish.A total of 211 integration units were sampled in the horizontal direction,and the integration values ranged from 901～1192247 m~2/nm~2,with a mean value of313548 m~2/nm~2.The NASC values of pacific saury in the vertical direction were mainly clustered within the 5-125m water layer,and the NASC was lower in deeper water layers,indicating that pacific saury were mainly distributed in the shallow water area within 125m.At the same time,the NASC values of saury at night were mainly distributed in the water layer of 5-35m,while those at daytime were distributed in the water layer of 65-125m.The overall NASC values at daytime were lower than those at night,indicating that there was a diurnal vertical migration pattern of pacific saury. |
