![]() It can make the learning process less effective and impact the recognition results. The object’s background is included so that the algorithm’s feature extraction is carried out both on the object and background. The conventional labeling technique is the most commonly used, but it has a weakness. There are three types of labeling used in this work: conventional, using landmarking, and a combination of conventional and landmarking. The labeling process is utilized during the training step. The training images and augmentations used in this work are summarized in Table 3. Furthermore, training images were also obtained by combining the original images with augmentations, as in scheme 1 as well. Then augmentation was applied to enrich the data with the same method as in scheme 1. Each fish that appears in the video was extracted by taking screenshots at three positions, as shown in Figure 3: when the fish appears in its entirety (a), at the exact center position (b), and just before the fish’s snout or tail hits the right-hand frame border to leave the frame (c). In scheme 2, the low-speed video was used and extracted into 188 static images. All images are then combined (original and augmented) and used as training images. So, every one of the 160 new images was obtained using vertical and horizontal flip techniques. Each augmentation technique was applied to each original image. The images were then augmented to enrich the data. Scheme 1 employed 160 static images of each fish from eight classes. Similar to the work, the approaches proposed in their papers were only for detecting fish without classifying them. They also used a fish dataset extracted from the Fish4Knowledge repository in their work. The GMM and Pixel-wise posteriors were proposed in, and then further developed by combining them with CNN. The combination of those methods enabled the robust detection and classification of fish, which was applied to the LifeCLEF 2015 benchmark dataset from the Fish4Knowledge repository and a dataset collected by the University of Western Australia (UWA) which was explained in detail by. The moving fish recognition in utilized Optical flow, GMM (Gaussian Mixture Models), and ResNet-50, then combined the output with YOLOv3. Using classic CNNs like these also has advantages when applied to other sectors, such as agriculture, or in other broad cases, such as detecting fine scratches. Even though their proposed method can reliably find and count fish objects in a variety of benthic backgrounds and lighting conditions, it is only used to find fish and not to classify them. For training and testing, they used a set of 18 underwater fish videos that were also recorded with a GoPro underwater camera. In, a multi-cascade object detection network with an ensemble of seven CNN components and two RPNs (Region Proposal Network) linked by sequentially jointly trained LSTMs (Long Short-Term Memory units) was performed. The best results were achieved in classifying 9 of the 20 types of fish that appear most often in the videos. They applied their proposed method to 116 underwater fish videos recorded using a GoPro underwater camera. The study in employed CNN (Convolutional Neural Network) to classify fish by training them with the number of species and their environments, such as reef bottoms and water. This study with a simple but effective method is expected to be a guide for automatically detecting, classifying, and sorting fish.įor moving fish recognition, many studies have been carried out. The proposed method was tested with videos of real fish running on a conveyor, which were put randomly in position and order at a speed of 505.08 m/h and could obtain an accuracy of 98.15%. This paper proposes an approach based on the recognition algorithm YOLOv4, optimized with a unique labeling technique. As far as the authors know, there has been no published work so far to detect and classify moving fish for the fish culture industry, especially for automatic sorting purposes based on the fish species using deep learning and machine vision. An automatic sorting system will help to tackle the challenges of increasing food demand and the threat of food scarcity in the future due to the continuing growth of the world population and the impact of global warming and climate change. Automatic fish recognition using deep learning and computer or machine vision is a key part of making the fish industry more productive through automation.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |