| Summary: | 碩士 === 國立高雄第一科技大學 === 電子工程系碩士班 === 106 === This paper proposes an Ornamental Fish Identification System Based on Image Retrieval. The system is an application example in aquaculture and agriculture. Living organisms in these fields are the main products, and it is very difficult to manage warehouses. In addition to counting and sorting, warehouse management also needs to identify the product, while the body shape and texture pattern of living organisms will change over time. It is more difficult to watch fish at high unit prices such as Koi, Arowana and Discus.
Take the Koi market for example, although the global market value is billions of dollars, the farming technology needs many years of accumulation and storage management difficulties, so it is difficult to expand the scale of operations, and the basis of business development is warehouse management. Even experienced fish farm operators need time to identify the koi identity, which may be misidentified, resulting in a misjudgment of the price of the fish. Traditionally based on RFID, it is necessary to capture koi individual implanted wafers, which is not only risky and time consuming.
Therefore, this paper proposes an ornamental fish recognition system based on image retrieval. It only takes 68ms to complete the action of recognizing the ornamental fish without harming the fish. The system has two subsystems, namely image pre-processing and machine. The learning engine, image pre-processing can receive the input source of the picture or the film, extract the image information of the fish and encode the texture pattern of the fish for subsequent identification of the identity. The machine learning engine recognizes the fish identity by the texture pattern code outputted by the front-end image pre-processing, and after successful recognition, the recognition module is re-trained to overcome the problem of the fish growing and the texture pattern changing with time.
This paper proposes the research contributions of two image processing, one is the Dynamic Background Model, which can effectively separate the front background, and the other is the Multi-Object Splitter, which can split the image close to multiple objects into multiple Single object image.
|
|---|
