|
...between to two weeks, the fish processing machinery operates around the clock (figure 1). Typically, fishes are brought on the boat and dropped into metal pockets that convey them through cleaning, cutting, and filleting machines. Anomalies, which must be detected at the beginning of the chain, include a fish of the wrong species or a damaged fish. Such anomalies must be rejected immediately. In addition, the presence of more than one fish in a pocket or the improper orientation of a single fish must be detected quickly to avoid jamming the cutting or filleting machines. This type of real-time inspection is not easy to deploy with conventional image-processing tools since the size, shape, and scales of fishes are difficult to model mathematically. Their features can also change depending on the location of the expedition as well as the season of the year. Figures 2-4 depict well-positioned and damaged fish. Finally, and most importantly, the inspection system must be very easy for crew members to use, since there is no room on board for a software programmer to fix a software bug or change an image-processing algorithm.
Several attempts have been done by Pisces Industries to solve this problem with traditional combinations of cameras, frame grabbers, PCs, and image-processing software. None of these attempts have led to a usable offshore system because of the high nonlinearity of the problem.
A neural network approach was the only possible way to deliver a system that could be both adaptive and trainable by the fishers themselves. A hardware neural network was the best way to deliver a reliable and fast system that featured both a small footprint and affordable cost. Typical fish species to be recognized include ill-defined herrings or mackerels.
Silicon Neural Network Justification
Due to the highly variable aspect of a fish, mathematical modeling was not an option. In addition all the "by catch" (which are random species) must be rejected, but random species cannot be learned due to the infinite number of shapes and textures. In order for the inspection system to operate properly, the concept of "uncertain" and "unknown" was critical. In addition, due to exacting conditions at sea, providing reliable continuous operation within a minimum space and without mechanical components (such as a PC fan) was mandatory. Speed was also essential--360 to 600 fishes per minute. Computerized statistical analysis was not a viable solution because of the need for a small footprint, high speed, and low cost. In real-life situations, the key is not necessarily to have the absolutely best classifier but instead to have a solution that can solve the problem at hand at a reasonable cost. In 2003 the ZISC (1) was the only available neural network chip that could meet these constraints. Its restricted coulomb energy (RCE) (Reilly, Cooper, and Elbaum 1988) feed-forward network offered a highly nonlinear classifier and allowed for unknown and uncertainty detection.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
Learning and Reinforcement
An important feature of the system was its ability to refine the learning "on board," since a fishing vessel can stay out...
NOTE: All illustrations and photos
have been removed from this article.
More articles from AI Magazine
Intelligent content discovery on the mobile Internet: experiences and ..., March 22, 2008
Custom DU--a web-based business user-driven automated underwriting sys..., March 22, 2008
A web-based agent challenges human experts on crosswords.(Report), March 22, 2008
Report on the eighteenth International Workshop on Principles of Diagn..., March 22, 2008
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
%20%20&position=1)