In the smart mariculture, timely and accurate predictions of water quality can help farmers take countermeasures before ecological environment deteriorates seriously. However, openness mariculture makes variation nonlinear, dynamic complex. Traditional methods face challenges in prediction accuracy generalization performance. To address these problems, an scheme is proposed for pH, temperature dissolved oxygen. First, we construct a new huge raw data set collected time series consisting...

The morphological features of fish, such as the body length, width, caudal peduncle pupil diameter, and eye diameter are very important indicators in smart mariculture. Therefore, accurate measurement is great significance. However, existing methods mainly rely on manual measurement, which operationally complex, low efficiency, high subjectivity. To address these issues, this paper proposes a scheme for segmenting fish image measuring based Mask R-CNN. Firstly, images acquired by home-made...

Infrared small target suffers from the lack of intrinsic features, context and samples. Conventional detection methods are usually unable to sufficiently effectively extract features infrared targets. Therefore, we propose a novel attention-based local contrast learning network (ALCL-Net). Considering scarcity targets, ResNet32, which enhances ability avoids problem that overwhelmed by background due too deep network. At same time, construct simplified bilinear interpolation attention module...

The normal growth of fishes is closely relevant to the density mariculture. It great significance accurately calculate breeding area specific sea from satellite remote sensing images. However, there are no reports about cage segmentation and detection based on images so far. And accurate cages faces challenges very large high-resolution Firstly, a new public mariculture data set built. Secondly, training augmented via sample variations improve robustness model. Then, for statistics,...

In the smart mariculture, batch testing of breeding traits is a key issue in improved fish varieties. The body length (BL), width (BW) and area (BA) features are important indicators. They great significance breeding, feeding classification. To accurately intelligently obtain morphological characteristic sizes actual scenes, data augmentation first used to greatly expand published dataset, thereby ensuring robustness training model. Then, an U-net segmentation measurement algorithm proposed,...

In smart mariculture, an automatic and accurate prediction of key water quality parameters is a significant challenge issue. This paper focuses on the pH temperature in parameters. Firstly, are preprocessed by improved method. Then, Pearson correlation coefficient method used to find between Finally, SRU (Simple Recurrent Unit) deep learning model establish for parameters, so as achieve prediction. Meanwhile, we also evaluate effect built RNN (Recurrent Neural Network) network. The...

Recently, feature relation learning has attracted extensive attention in cross-spectral image patch matching. However, most methods can only extract shallow relations and are accompanied by the loss of useful discriminative features or introduction disturbing features. Although latest multi-branch difference network relatively sufficiently features, structure it adopts a large number parameters. Therefore, we propose novel two-branch interaction (FIL-Net). Specifically, idea for matching is...

Recently, infrared small target detection has attracted extensive attention. However, due to the size and lack of intrinsic features targets, existing methods generally have problem inaccurate edge positioning is easily submerged by background. Therefore, we propose an innovative gradient-guided learning network (GGL-Net). Specifically, are first explore introduction gradient magnitude images into deep learning-based method, which conducive emphasizing details alleviating targets. On this...

Cross-spectral image patch matching is still challenging due to significant nonlinear differences between patches. Recently, methods based on feature relation learning have attracted increasing attention and achieved good performance. However, we find that the metric difference cannot comprehensively effectively extract useful discriminative information pairs by only adopting two branches network structure. Therefore, propose a novel multi-branch (MFD-Net). Specifically, build parallel...

In aquaculture, using high-resolution SAR images to precisely segment offshore farms is helpful for reasonable layout planning and statistics of breeding density. However, conventional segmentation methods tend have low accuracy slow inference speed. Therefore, we propose a novel, precise fast scheme in based on model fusion half-precision parallel inference. Specifically, several new high-performance improved UNet++ reasonably fuse the test results. At same time, simulated annealing...

Recently, cross-spectral image patch matching methods based on feature difference aggregation have achieved excellent performance, but they introduce a large number of parameters, limit speed and poor scalability. At the same time, only using learning to extract differential features will lead loss consistent between patches. Therefore, we construct novel four-branch efficient relation network (EFR-Net) without aggregation. Specifically, new strategy is proposed, which reasonably combines...

Deep learning has contributed to the rapid development of building extraction tasks from remote sensing(RS) images. Existing models typically leverage a segmentation-head predict results, where multi-channel feature maps extracted by network are directly output as single-channel predictions. However, it is rarely noticed that this process results in loss features, which can lead incomplete smaller buildings. Besides, boundary-blurring also common problem task. Therefore, letter, we propose...

Infrared small target detection faces the problem that it is difficult to effectively separate background and target. Existing deep learning-based methods focus on appearance features ignore high-frequency directional features. Therefore, we propose a multi-scale direction-aware network (MSDA-Net), which first attempt integrate of infrared targets as domain prior knowledge into neural networks. Specifically, an innovative multi-directional feature awareness (MDFA) module constructed, fully...

Recently, deep learning-based single-frame infrared small target (SIRST) detection technology has made significant progress. However, existing methods are often optimized for a fixed image resolution, single wavelength, or specific imaging system, limiting their breadth and flexibility in practical applications. Therefore, we propose refined scheme based on an adjustable sensitivity (AS) strategy multi-scale fusion. Specifically, model fusion framework direction-aware network (MSDA-Net) is...

Limited by equipment limitations and the lack of target intrinsic features, existing infrared small detection methods have difficulty meeting actual comprehensive performance requirements. Therefore, we propose an innovative lightweight robust network (LR-Net), which abandons complex structure achieves effective balance between accuracy resource consumption. Specifically, to ensure robustness, on one hand, construct a feature extraction attention (LFEA) module, can fully extract features...

Fish behavior can be used as an important indicator of the water quality in mariculture. Therefore, changes fish reflect a timely and effective manner. In order to track behavior, target location tracking algorithm based on multi-domain deep convolutional neural network is proposed. Firstly, cross-entropy loss function distinguish background each domain network, optimized by stochastic gradient descent method (SGD) find local minimum function. Then hard example mining resolve positive...

In recent years, data-driven deep networks have demonstrated remarkable detection performance for infrared small targets. However, continuously increasing the depth of neural to enhance has proven impractical. Consequently, integration prior physical knowledge related targets within become crucial. It aims improve models’ awareness inherent characteristics. this paper, we propose a novel dual-domain prior-driven network (DPDNet) small-target detection. Our method integrates advantages both...

Recently, single-frame infrared small target (SIRST) detection with single point supervision has drawn wide-spread attention. However, the latest label evolution (LESPS) framework suffers from instability, excessive evolution, and difficulty in exerting embedded network performance. Therefore, we construct a Progressive Active Learning (PAL) framework. Specifically, inspired by organisms gradually adapting to their environment continuously accumulating knowledge, propose an innovative...

Recently, infrared small target detection with single-point supervision has attracted extensive attention. However, the accuracy of existing methods difficulty meeting actual needs. Therefore, we propose an innovative refined scheme supervision, which excellent segmentation and rate. Specifically, introduce label evolution single point (LESPS) framework explore performance various networks based on this framework. Meanwhile, to improve comprehensive performance, construct a complete...