Initial Research to Predict The Length and Weight of Nile Tilapia Using ImageAI

One of the processes of cultivating Nile Tilapia requires measuring the mass and size of the fish. These measurements are carried out manually, which triggers stress, affects fish growth, and produces fish of lower quality. This system was created to help farmers improve the quality of fish growth rates. Initial research on the fish length and weight prediction system that has been designed goes through several stages, namely collecting actual fish length and weight data, collecting fish image datasets, labeling fish object datasets, and training fish objects with a dataset of 993 images using the ImageAI library, which compares the performance between the two architectures used, namely YOLO (You Only Look Once) V3 and Tiny YOLO V3. The result is that the speed of the Tiny YOLO V3 training process is around 14 hours faster than YOLO V3 around 52 hours. However, the detection results from YOLO V3 have better accuracy, namely 93 percent of the 15 datasets tested, than Tiny YOLO V3, which is only 80 percent. The fish image dataset was taken with a distance between the table and the camera, namely 33 cm. Measuring the length and weight of the fish with the height and width of the fish image pixels has a relationship, namely the longer and wider the fish, the longer and wider the fish image pixel size will be if the detection results are accurate. The conclusion is that the YOLO V3 architecture was chosen to be used for detecting fish objects; apart from that, there are results of Nile Tilapia growth patterns from the formulation of the actual relationship between the length and weight of the fish. There are also height and width values for the image resulting from pixel measurements of the fish image. Thus making this initial research valid and can be continued to complete the research of predicting the length and weight of fish automatically by creating a calculation formula.