Conventional feeders can achieve timed and quantitative feeding, but they cannot optimize feeding strategies on the basis of actual aquaculture conditions. This study evaluated the feeding intensity of largemouth bass and developed an intelligent feeder to achieve efficient and precise feeding. A mobile feeding system was built by designing and simulating the structure of the data acquisition, control, feeding power, storage, and mobile modules of the feeder. The surface water pressure signals during largemouth bass feeding were collected through pressure sensors and analyzed, and the feeding intensity was classified into three levels: strong, weak, and none. Signal features were extracted to construct a dataset and input into five machine learning models for optimal parameter tuning. The precision, recall, F1 score, and average accuracy of the random forest model were 96.2%, 95.5%, 95.6%, and 93.4%, respectively. The YOLOv5 model was adopted to detect remaining feed on the water surface. The feeding system was designed to enable the feeder to automatically track and provide feed into the tank. Experiments were conducted on the intelligent feeding system, with the feed residue rate as the indicator of the practicality of the feeding system. Verification experiments were also performed on eight tanks, and the average feed residue rate was less than 3%, proving that the feeding system has good practicality in actual aquaculture environments.

This study aims to realize the head, tail and ventral dorsal orientation of fish bodies in conveyors using machine vision technology. The Scomber japonicus was taken as the object of study. The morphological and physical characteristics of the fish were used to design the mackerel body lifting device, the fish body separation device, the fish body head, fishtail and ventral dorsal directional conveying device, the fish body return conveying device, and the directional control system to form the mackerel body directional arrangement conveying device. The fish body lifting device included a spacer conveyor belt and brush rollers to separate and convey the fish upwards; the fish body separation device included a fish sliding mechanism and brush rollers to guide the fish from the lifting device to the fish separating device, in order to separate and convey the fish; the fish body head, tail and ventral dorsal directional conveying device consisted of an image acquisition device, head- and tail- and ventral-dorsal directional actuator to convey the fish in a set head, tail and ventral dorsal directional direction. The fish return conveyor was used to return the fish with the wrong head and tail orientation, in order to the fish lifting device for reorientation. The mackerel dataset was created. The image data was then enhanced with brightness enhancement, Gaussian noise and rotation to enrich the number of training datasets. The annotation of the image data was finally realized. YOLOv5s was selected as the fish head, tail and ventral and dorsal orientation detection model, with an accuracy of 99.76%, a recall of 99.59%, and an average detection accuracy value of 99.5%. The head, tail and ventral and dorsal orientation of the fish were detected in real time on the device. The orientation control system consisted of an Arduino UNO controller, a computer and detection program, a CMOS industrial camera, a head and tail orientation actuator cylinder, a ventral and dorsal orientation actuator cylinder, a solenoid valve, an optocoupler isolated relay, a photoelectric sensor and a power supply. The orientation control system controlled the head, tail and ventral dorsal orientation of the mackerel using the identification results of the fish head, tail and ventral dorsal orientation model, and finally realized that the fish were transported forward in a certain head, tail and ventral dorsal orientation. As such, a prototype of the fish body directional arrangement conveying device was tested to verify the conveying effect of the fish body lifting device at different conveying speeds using the single fish body lifting success rate as the evaluation index. At the same time, the directional success rates of the fish body head and tail, as well as ventral and dorsal were used as evaluation indexes, and the conveying speed of the fish body lifting device, the conveying speed of the fish body separation conveying device, the fish body head, tail and ventral and dorsal directional conveying. A series of tests were performed on the effect of the directional transport of fish in a directional arrangement transport device. The test results show that the fish lifting device effectively separated and lifted the fish at different conveying speeds, and there was no overlapping of the fish upwards. The directional conveying speed of the fish body reached 15 pieces/min when the conveying speed of the fish lifting device was 0.05 m/s, the conveying speed of the fish separating device was 0.45 m/s, and the conveying speed of the fish head, tail and ventral dorsal orientation device was 0.60 m/s. The finding can offer a strong reference for the development of freshwater fish orientation conveyors.

An accurate and rapid identification of species is one of the most important parts of freshwater fish pre-treatment. However, deep learning-based freshwater fish classification cannot quantitatively describe the feature parameters of fish, due to the complicated feature extraction. This study aims to propose a deep learning-based fish feature point detection and freshwater fish classification. Firstly, an image acquisition device was built to acquire the images of five freshwater fish species: bream, bighead carp, grass carp, silver carp, and common carp. Secondly, the data enhancement was performed on the original images. LabelMe software was also used to label 20 feature points of the fish body, in order to construct a freshwater fish feature point detection dataset. The feature point of freshwater fish was detected using an improved AlexNet feature point detection model with the convolutional neural networks (CNN). The improved AlexNet model also adjusted the network structure of the traditional model, in order to accommodate the feature point detection task by the modified convolutional kernel size, the removal of the local response normalization layer, the addition of batch normalization layer, and the replacement of the loss function. The Euclidean distance between feature points was extracted as the feature value using the coordinates of the feature points. Feature vectors were constructed. Fisher discriminant analysis was used to identify the species of freshwater fish. A series of ablation experiments were conducted on the improved AlexNet model to clarify the impact of different improvement schemes on model performance. The results showed that the improved schemes were beneficial to improve the training efficiency and performance of the model. The average value of normalized mean error (NME) of the improved AlexNet model on the test set was 0.0099, the failure rate (FR) at thresholds of 0.02 and 0.03 were 2.50% and 0.83%, and the average detection time was 0.037 s, indicating the better accuracy and error distribution. The number of parameters and floating-point operations per second (FLOPs) of the improved AlexNet model were smaller than those of the VGG16 model. The test set detection was also better than that of the VGG16 model, compared with different feature extraction models. Although the number of parameters and FLOPs of the improved AlexNet model was larger, the average NME and the FR were lower than that of the MobileNetV2 model. The recognition accuracy of 98.0% was achieved in the freshwater fish species recognition using the model and Fisher discriminant analysis for five freshwater fish species. The average recognition time of a single image was 0.368 s. The feature point detection model can be expected to detect the feature points of five species of freshwater fish with high accuracy. The finding can also facilitate freshwater fish species identification, size detection, and fish segmentation. The freshwater fish species identification using feature point detection and Fisher's discriminant analysis can lay the foundation for the development of an automated freshwater fish classification device.

The widespread mortality has often occurred in aquaculture systems in recent years, due to the high stocking densities. The presence of a few dead fish in tanks can also lead to the mass death of the overall population. In this study, the accurate and rapid detection of the dead fish was proposed to combine with the key point correction, according to their underwater posture. A series of experiments was conducted on the largemouth bass (Micropterus salmonids) under the culture mode of the round culture barrels. A dataset was then captured to detect the underwater dead fish. The MLP-Block (Multi-Layer Perceptron) was combined with the MSPCA (Multi-Path Coordinate Attention) mechanism to enhance the dynamic convolution. The MLPNet-Pose algorithm was used to group the decoupling head for the fused features from the path aggregation network. Both target detection and key point detection outputs were realized after feature fusion. Furthermore, a key point correction was applied to classify the posture features of the underwater fish. Thereby, the accurate identification of dead fish was achieved after correction. The posture behavior of the fish after death was obtained as the key indicators for detection, such as abnormal floating or reduced movement. A dataset was specifically constructed to detect the dead fish. The target detection was also integrated with the key point detection. The high precision was obtained to differentiate between live and dead fish. The MLPNet-Pose with the dynamic convolution was enhanced by a multi-path coordinate attention mechanism. Some subtle differences in posture between live and dead fish were effectively captured and then processed after optimization. Moreover, the decoupling head improved the efficient fusion of features and the accuracy of the detection. The key point correction was used to refine the classification of the fish postures. There were some variations in the fish's orientation and movement. The reliable identification of the dead fish was realized under complex and dynamic underwater environments. Experimental results show that the improved algorithm was achieved in the detection accuracies of 99.1% for the live fish and 96.0% for the dead fish on the test dataset. The key point detection demonstrated high precision, high speed, and lower parameter count, particularly suitable for real-time applications. A solid theoretical and technical solution was offered to identify the key point and underwater dead fish for the optimal feeding strategy. In summary, the improved detection of the key point can offer a highly accurate and efficient solution to identify the underwater dead fish in aquaculture systems. The target detection and key point correction were combined to operate effectively in real time, particularly for the robust and scalable approach in modern aquaculture. The findings can also provide valuable technical insights for the detection of the dead fish, feeding strategy, and underwater fish monitoring.

A homogeneous and stable flow field is required to promote the healthy growth of the fish in the recirculating aquaculture systems. Among them, the structure of the inlet pipes can dominate the water inflow patterns and the distribution of the flow field in the aquaculture ponds. At the same time, there is also a great influence on the growth of the fish and the removal of suspended particulate matter, such as the uneaten feed and the feces. This study aims to investigate the impacts of the different structures of the inlet pipes on the flow field distribution in circular aquaculture ponds. The optimal structural parameters of the inlet pipes were determined for the uniform distribution of the flow field. A numerical model was established for the circular aquaculture pond using computational fluid dynamics (CFD) software. A series of simulations were then conducted to verify the structural parameters, including the number of the inlet pipes, the ejection angle of each inlet pipe, and the number of the jet holes on the inlet pipes. The uniformity of the flow field was evaluated, according to the uniformity coefficient of velocity distribution U50 as an assessment indicator. Orthogonal experiments were designed to obtain the optimal combination for the structural parameters of the inlet pipes in the circular aquaculture pond. Simultaneously, a recirculating aquaculture system was constructed in accordance with the optimal structure parameters of the inlet pipes. An aquaculture experiment was carried out on Micropterus salmoides, in order to verify the optimization. The flow field was then evaluated to monitor the water quality and the growth indicators of Micropterus salmoides samples in the aquaculture system. The experimental results demonstrate that the most uniform distribution of the flow field was achieved, when the circular aquaculture pond was equipped with three inlet pipes arranged at 120°, with six jet holes on each single inlet pipe and a jet angle of 45° for the inlet pipe. Three groups of Micropterus salmoides aquaculture experiments were conducted with different densities at the water flow velocity of 0.1 m/s. The concentration values of the nitrite and ammonia nitrogen were less than or equal to 0.2 and 0.3mg/L, respectively in the aquaculture pond. The turbidity values of the water were all lower than 0.3 NTU (nephelometric turbidity unit), indicating the high water quality and the low level of the suspended particles. The lowest feed coefficient of Micropterus salmoides samples with the three different density groups was 1.06, while the highest weight gain rate was 161.03%, and the highest survival rate was 100%. The flow field contributed to the weight gain rate of fish for the favorable survival rate of the fish population. The feed coefficient of fish was also reduced to facilitate the healthy growth of fish. The optimal structure of the inlet pipes was conducive to the uniform and stable flow field in a circular aquaculture pond. The suspended particulate matter was effectively discharged to maintain the water quality of the aquaculture water body. The finding can provide theoretical guidance to optimize the structures of the inlet pipes for the uniform flow field in a recirculating aquaculture system.