Coral Fish Shoal Detection from Acoustic Echograms

An algorithm was developed that processes data from a single-beam echosounder and GPS system to detect and analyze fish shoals. The proposed algorithm provides the ability to identify regions within an echogram that contain fish and statistically analyze the shoals. The echosounder used is a 50 kHz/200 kHz dual-frequency unit manufactured by Lowrance and was selected for its low-cost, off- the-shelf availability, and portability. The selected unit saves the data onto removable memory cards allowing for easy transfer and offline processing. The development of this algorithm faced several challenges: translating the data to a common format, bottom identification, noise rejection, and fish school recognition. The data is saved by the echosounder in a unique binary format but is parsed into a more manageable data structure by MATLAB software developed at Portland State University. The software, called EchoMap, was developed under sponsorship by The Nature Conservancy (TNC) in support of their marine habitat monitoring activities. EchoMap processes and interprets the echosounder data to generate coral reef profiles and to estimate the associated fish populations. This paper discusses the algorithm developed for fish shoal detection and cross-sectional shoal area estimation. Abundance and density of the fish shoals is assumed to be directly related to the area of the identified shoal. Shoals are segmented with a dynamic contouring method and typically have non-uniform shape. The area calculation was improved from the standard Riemann sum approach by using Green´s theorem to translate the line integral of a closed level set contour into a surface integral. The advantages of using Green´s theorem are higher accuracy and efficiency in estimating contour areas. Preliminary results of applying the detection algorithm to coral reef echograms provided by TNC show promise in the ability to identify fish shoals. Results were compared to detections made by experts in the field fr- om echograms as well as the real-time Lowrance Fish I.D. algorithm. Applying a priori knowledge of fish characteristics may also allow species identification using the calculated descriptors.