Automatic fish recognition using deep learning and computer or machine vision is a key part of making the fish industry more productive through automation. 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. 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. This paper proposes an approach based on the recognition algorithm YOLOv4, optimized with a unique labeling technique. 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 study with a simple but effective method is expected to be a guide for automatically detecting, classifying, and sorting fish.

Food scarcity is an issue of concern due to the continued growth of the human population and the threat of global warming and climate change. Increasing food production is expected to meet the challenges of food needs that will continue to increase in the future. Automation is one of the solutions to increase food productivity, including in the aquaculture industry, where fish recognition is essential to support it. This paper presents fish recognition using YOLO version 4 (YOLOv4) on the "Fish-Pak" dataset, which contains six species of identical and structurally damaged fish, both of which are characteristics of fish processed in the aquaculture industry. Data augmentation was generated to meet the validation criteria and improve the data balance between classes. For fish images on a conveyor, flip, rotation, and translation augmentation techniques are appropriate. YOLOv4 was applied to the whole fish body and then combined with several techniques to determine the impact on the accuracy of the results. These techniques include landmarking, subclassing, adding scale data, adding head data, and class elimination. Performance for each model was evaluated with a confusion matrix, and analysis of the impact of the combination of these techniques was also reviewed. From the experimental test results, the accuracy of YOLOv4 for the whole fish body is only 43.01 %. The result rose to 72.65 % with the landmarking technique, then rose to 76.64 % with the subclassing technique, and finally rose to 77.42 % by adding scale data. The accuracy did not improve to 76.47 % by adding head data, and the accuracy rose to 98.75 % with the class elimination technique. The final result was excellent and acceptable

The formation of fouling deposits on heat exchanger surfaces is one of the major concerns in thermal processes. The fouling behavior of food materials is complex, and its mechanism remains, in general, unclear. This study was aimed at developing a predictive model for soymilk fouling deposit formed on heated surfaces using dimensional analysis. Relevant variables affecting fouling deposit mass could be grouped into six dimensionless terms using Buckingham’s pi-theorem. Experimental data were obtained from a lab-scale plate heat exchanger. A simple model developed using the experimental data under the process conditions with the product inlet temperature, the product outlet temperature, and plate surface temperature in the ranges of 50–55 °C, 65–70 °C, and 70–85 °C, respectively, exhibited a good performance in the prediction of soymilk fouled mass. The correlation coefficient between the predicted and experimental values of fouled mass was 0.97 with an average relative error of 9.03%. Within the ranges of product inlet temperature and plate surfaces temperature studied, this model offers an opportunity to estimate soymilk fouling mass with acceptable accuracy.

The advancement of automation in the fish industry, a critical segment of the food sector, has become increasingly relevant in light of the growing global population and the impacts of climate change and global warming. Enhancing productivity through automation is essential to mitigate the looming threat of food scarcity. In this context, automatic fish classification using computer vision has garnered significant attention, with various studies exploring both complex and simple approaches. While complex methods have shown promising results, simpler approaches often fall short in performance. This study proposes a simple yet effective method that highlights key distinguishing features of fish—namely, body shape and scale patterns—for species classification. The Lanczos resampling technique is employed to crop, resize, and focus on the features, enabling a lightweight deep learning model to effectively learn and classify fish species. With the right conceptual framework, appropriate feature extraction techniques, and an efficient deep learning architecture, the proposed method addresses the classification challenge in a straightforward yet effective manner. Experimental evaluations using the Fish‐Pak dataset, comprising six aquaculture fish species, and the KMITL Fish dataset, containing eight species, demonstrate the effectiveness of the method, achieving accuracy rates of 97.16% and 98.59%, respectively.