Image Processing Reference
In-Depth Information
1 Introduction
Generally, small farms use a manual measuring and counting process when cultivating rain-
bow trout [ 1 - 4 ]. There are many reasons to perform such classification, but the most important
are to feed the trout according to its size and to avoid cannibalism into the tanks [ 4 ] . Problems
when doing a manual classification are, indistinctively, the stress and physical damage causes
to the specimen when manipulated by the farmer. Moreover, we believe that this classification
approach is not accurate, where the trout is taken from the water using a net and visually the
farmer decide whether or not the trout should be changed to another tank.
Mexico, as well as many other countries in the world, has large hydric areas, which are ideal
for aquaculture [ 5 , 6 ]. Taking advantage of both, its altitude and natural water resources, the
State of Mexico (Mexico) has particular interest in increasing the trout's production as a sus-
tainability and economic strategy for local small farmers [ 7 ] . Hence, this is a good opportunity
to integrate technology to optimize the trout's production in this region.
That is the reason which motives our research interest in the field, where we have accom-
plished some results, including a research project [ 3 ] and a couple of bachelor in science dis-
sertations [ 1 , 2 ] . In this paper we robustly evaluate our experimental procedure to measure
rainbow trout [ 8 ] in a small farm located in the Valley of Toluca, Mexico [ 9 ] , where we have
observed a manual classification process as illustrated in Figure 1 .
FIGURE 1 Manual measuring-classification process generally done in small farms in central
Mexico. Note that this small farms use lined earth tanks.
Therefore, robust experimental results are presented in this publication by using our statist-
ical system [ 8 ] and a state-of-the-art rainbow trout image database specially collected for this
article. These data corpus were collected by capturing 20 images for each of 30 specimens per
size (fry, fingerling, and table-fish), counting 1800 rainbow trout images.
Some related work is observed in the literature. Hsieh et al. [ 10 ] proposed a technique to
measure dead tuna ish using a colour patern. In this work, the ish length is estimated by
proportional relationship between the fish body pixel length and an image reference scale.
Ibrahim and Wang [ 11 ] measure four dead fish classes by constructing a central line along the
fish body from horizontal and vertical views of the fish's body. Finally, a commercial counting
and measuring system is observed in Vaki System [ 12 ] ; however, there is no further informa-
tion about its classification procedure.
The rest of this article is as follows. First, Section 2 describes our novel prototype designed
for this research. Then, Section 3 introduces our statistical measuring approach. After that,
Section 4 details our experimental framework. Next, Section 5 shows our performance evalu-
ation. Finally, Section 6 concludes this article and draws some venues for our future work.
 
Search WWH ::




Custom Search