Lobster is an invertebrate animal belonging to the phylum Arthropoda that lives in water. It is a type of shrimp with great economic value in Indonesia. Some lobstermen still use conventional methods to identify lobster species, such as visual methods to distinguish sand lobster from batik lobster. The proposed method employs YOLOv7 to perform species identification on video footage. In addition to being accurate, the proposed method is also efficient. The detection speed ranged from 7.9ms to 14.1ms per frame, and the inference speed ranged from 0.8ms to 2ms per frame. This means that the model can detect lobster species in real time, which is important for lobstermen who need to identify lobster species quickly and accurately. To evaluate the performance of the proposed method, the researchers trained the YOLOv7 model on a dataset of over 5,000 images of lobster from two different species. The model was then tested on a held-out test set 2-minute video length of lobster. The researchers also evaluated the speed of the proposed method on a 2-minute video of lobster with 7562 frames. It took 3 minutes and 15 seconds to detect lobster in the video. This is a relatively fast detection speed, considering that the model was able to achieve a mAP of 99.6%. Overall, the results of this study showed that the use of YOLOv7 is successful and efficient. With this technology, lobstermen can effectively distinguish between these two species, enhancing the profitability and sustainability of lobster farming.

Link to Publication : https://ieeexplore.ieee.org/document/10458215