User menu
We have been developing three different types of ICT or AI based data collection andtransmission systems for fisheries related information & sea conditions using mobile devices.The outline of three systems are described as below:
(1) GPS data logger
The GPS data logger is the real-time data collection and transmission system for fisheriesrelated information and sea conditions using a tablet connected to GPS, sensor and echosounder on small coastal fishing vessels and has the following five functions:
(i) to collectexact vessel track lines every second,
(ii) to collect & enter catch by species for each set by fishers,
(iii) to collect 3-D gear locations and sea temperature by small wireless sensors attached to the gears and also to collect the bottom depth data from the echo sounder,
(iv) to transmit these data to the GPS data logger to produce automatic display of depth and sea temperature profiles for each set, and
(v) to transmit all data to the cloud server and update the database for users to utilize.
The collected information can be used to implement the following four important tasks:
(i) real-time monitoring of fisheries related information and sea conditions,
(ii) quick identification of good fishing grounds by sharing data from all vessels using GPS data loggers for higher catch and lower fuel costs (to get more profits),
(iii) pinpoint forecasting for good fishing grounds using accumulated information, and
(v) reliable resource managements (e.g., to establish fine scale closed areas and periods).
Our GPS data loggers are currently used by 370 boats with different types of gears including tuna troll fisheries (yellowfin, skipjack, and neritic tuna), hand harpoon swordfish fisheries, trawl, purse seine, gillnet and squid fisheries. In the research area, our GPS data logger is utilized by Pacific bluefin tuna recruitment monitoring survey. At landing sites, our system is used also with smartphones (no GPS).
We are currently developing two additional AI based systems, i.e., fish species identification and fish size measurements using Neural Network (NN) which outlines are described as follows:
(2) Fish species identification by NN
We apply AI (Neural Network) to identify fish species using images of catch picture (including many species) taken by smartphones or tablets. Results of our preliminary trials based on NN learnings suggest, (i) max. 80 fish species can be identified, (ii) species can be also identified from parts of the fish body, and (iii) accuracy is subject to quantity of learning and quality of images. We are also planning to develop a tuna-specific model to identify between juvenile yellowfin vs. bigeye tuna, frigate vs. bullet tuna, and billfish species from the body picture (images).
(3) Fish size estimation by NN
A few preliminary studies (including tuna) suggest (i) there are linear relationships between eye size and fish body size with various levels of correlations by species, and (ii) eye size from the image is robust (accurate) to measure as it is not affected by photo angles unlike the longer object (e.g. fork length).
Following two suggestions, we are currently developing a system to estimate fish size from eye size using five steps : (i) to estimate relations between fish eye vs. body size (fork length) by real measurements, (ii) to identify species from image (catch photo) by NN as explained previously, (iii) to let NN to learn locations of eyes from images of the fish, (iv) to measure eye size (pixel) in the image of the fish, (v) to estimate fish size using the linear relation. SE, CV & r2 (in the linear relation) and classification errors (NN) will be provided for users to understand levels of uncertainties.
Our goal
Based on three systems outlined above, our final goal is to build an integrated system by combining all of three into one GPS data logger. With this integrated system, we can (i) collect fisheries related and sea condition information, (ii) identify species from images of multiple fish species pictures taken by tablet or smartphone and (iii) to estimate fish size by measuring the eye size from eye images of fish identified and by substituting it to pre-learnt relation between fish eye and size.
