Automated detection and classification of southern African Roman seabream using mask R-CNN

The availability of relatively cheap, high-resolution digital cameras has led to an exponential increase in the capture of natural environments and their inhabitants. Video-based surveys are particularly useful in the underwater domain where observation by humans can be expensive, dangerous, inaccessible, or destructive to the natural environment. However, a large majority of marine data has never gone through analysis by human experts – a process that is slow, expensive, and not scalable. We test a Mask R-CNN object detection framework for the automated localisation, classification, counting and tracking of fish in unconstrained underwater environments. We present a novel, labelled image dataset of roman seabream (Chrysoblephus laticeps), a fish species endemic to Southern Africa, to train and validate the accuracy of our model. The Mask R-CNN model accurately detected and classified roman seabream on the training dataset (mAP50 = 80.29%), validation dataset (mAP50 = 80.35%), as well as on previously unseen footage (test dataset) (mAP50 = 81.45%). The fact that the model performs well on previously unseen data suggests that it is capable of generalising to new streams of data not included in this research.     

YOLO-Fish: A robust fish detection model to detect fish in realistic underwater environment

Over the last few years, several research works have been performed to monitor fish in the underwater environment aimed for marine research, understanding ocean geography, and primarily for sustainable fisheries. Automating fish identification is very helpful, considering the time and cost of the manual process. However, it can be challenging to differentiate fish from the seabed and fish types from each other due to environmental challenges like low illumination, complex background, high variation in luminosity, free movement of fish, and high diversity of fish species. In this paper, we propose YOLO-Fish, a deep learning based fish detection model. We have proposed two models, YOLO-Fish-1 and YOLO-Fish-2. YOLO-Fish-1 enhances YOLOv3 by fixing the issue of upsampling step sizes of to reduce the misdetection of tiny fish. YOLO-Fish-2 further improves the model by adding Spatial Pyramid Pooling to the first model to add the capability to detect fish appearance in those dynamic environments. To test the models, we introduce two datasets: DeepFish and OzFish. The DeepFish dataset contains around 15k bounding box annotations across 4505 images, where images belong to 20 different fish habitats. The OzFish is another dataset comprised of about 43k bounding box annotations of wide varieties of fish across around 1800 images. YOLO-Fish1 and YOLO-Fish2 achieved average precision of 76.56% and 75.70%, respectively for fish detection in unconstrained real-world marine environments, which is significantly better than YOLOv3. Both of these models are lightweight compared to recent versions of YOLO like YOLOv4, yet the performances are very similar.     

Posidonia australis seed predation in seagrass habitats of Two Peoples Bay,Western Australia

Posidonia australis seed predation experiments conducted in three seagrass habitats (P. australis, Posidonia sinuosa, Halophila ovalis) and bare sand in Two Peoples Bay, Western Australia, showed higher predation rates in seagrass than bare sand, supporting general conclusions from two previous predation studies in Western Australia. However, much higher rates were noted in H. ovalis, compared to previous observations of very low rates in H. ovalis on Rottnest Island, Western Australia. We attribute these differences to gammaridean amphipods (family Lysianassidae) that were present in a detrital layer within the H. ovalis in Two Peoples Bay. Our data from Two Peoples Bay continues to add to the growing body of information showing high seed predation rates in most seagrass habitats by a diverse group of crustacean species.     

Seagrass Proliferation Precedes Mortality during Hypo-Salinity Events: A Stress-Induced Morphometric Response

Halophytes, such as seagrasses, predominantly form habitats in coastal and estuarine areas. These habitats can be seasonally exposed to hypo-salinity events during watershed runoff exposing them to dramatic salinity shifts and osmotic shock. The manifestation of this osmotic shock on seagrass morphology and phenology was tested in three Indo-Pacific seagrass species, Halophila ovalis, Halodule uninervis and Zostera muelleri, to hypo-salinity ranging from 3 to 36 PSU at 3 PSU increments for 10 weeks. All three species had broad salinity tolerance but demonstrated a moderate hypo-salinity stress response – analogous to a stress induced morphometric response (SIMR). Shoot proliferation occurred at salinities <30 PSU, with the largest increases, up to 400% increase in shoot density, occurring at the sub-lethal salinities <15 PSU, with the specific salinity associated with peak shoot density being variable among species. Resources were not diverted away from leaf growth or shoot development to support the new shoot production. However, at sub-lethal salinities where shoots proliferated, flowering was severely reduced for H. ovalis, the only species to flower during this experiment, demonstrating a diversion of resources away from sexual reproduction to support the investment in new shoots. This SIMR response preceded mortality, which occurred at 3 PSU for H. ovalis and 6 PSU for H. uninervis, while complete mortality was not reached for Z. muelleri. This is the first study to identify a SIMR in seagrasses, being detectable due to the fine resolution of salinity treatments tested. The detection of SIMR demonstrates the need for caution in interpreting in-situ changes in shoot density as shoot proliferation could be interpreted as a healthy or positive plant response to environmental conditions, when in fact it could signal pre-mortality stress.     

Development of a ‘sediment-stress’ functional-level indicator for the seagrass Halophila ovalis

Understanding mechanistic relationships between seagrass and their environmental stressors should be considered for effective management of estuaries and may inform on why change has occurred. We aimed to develop indicators for seagrass health in response to sediment conditions for the Swan-Canning Estuary, south-west Australia. This article describes the development of a new sediment-stress indicator, relating aspects of seagrass productivity with sediment sulfur dynamics. Sulfur stable isotope ratio and total sulfur were measured monthly within the roots, rhizomes and leaves of Halophila ovalis, and significantly varied across sites and months. The growth of seagrass over the summer months appeared restricted by sediment condition, with growth of seagrass lower when sediment derived sulfur and/or total sulfur within rhizome of leaf tissues was higher. H. ovalis appeared quite tolerant of sulfide intrusion within the root compartment, but growth was compromised when sulfide breached the root–rhizome barrier. The tightest correlation between potential sulfur metrics and seagrass growth was observed for the ratio (δ³⁴S_leaf + 30)/(TS_leaf), and it is this ratio that we propose may be a useful sediment-stress indicator for seagrass. The study also highlights that sediment condition needs to be considered at the meadow scale.     

Automatic strawberry leaf scorch severity estimation via faster R-CNN and few-shot learning

Using deep learning to estimate strawberry leaf scorch severity often achieves unsatisfactory results when a strawberry leaf image contains complex background information or multi-class diseased leaves and the number of annotated strawberry leaf images is limited. To solve these issues, in this paper, we propose a two-stage method including object detection and few-shot learning to estimate strawberry leaf scorch severity. In the first stage, Faster R-CNN is used to mark the location of strawberry leaf patches, where each single strawberry leaf patch is clipped from original strawberry leaf images to compose a new strawberry leaf patch dataset. In the second stage, the Siamese network trained on the new strawberry leaf patch dataset is used to identify the strawberry leaf patches and then estimate the severity of the original strawberry leaf scorch images according to the multi-instance learning concept. Experimental results from the first stage show that Faster R-CNN achieves better mAP in strawberry leaf patch detection than other object detection networks, at 94.56%. Results from the second stage reveal that the Siamese network achieves an accuracy of 96.67% in the identification of strawberry disease leaf patches, which is higher than the Prototype network. Comprehensive experimental results indicate that compared with other state-of-the-art models, our proposed two-stage method comprising the Faster R-CNN (VGG16) and Siamese networks achieves the highest estimation accuracy of 96.67%. Moreover, our trained two-stage model achieves an estimation accuracy of 88.83% on a new dataset containing 60 strawberry leaf images taken in the field, which indicates its excellent generalization ability.     

WilDect-YOLO: An efficient and robust computer vision-based accurate object localization model for automated endangered wildlife detection

Objective. With climatic instability, various ecological disturbances, and human actions threaten the existence of various endangered wildlife species. Therefore, an up-to-date accurate and detailed detection process plays an important role in protecting biodiversity losses, conservation, and ecosystem management. Current state-of-the-art wildlife detection models, however, often lack superior feature extraction capability in complex environments, limiting the development of accurate and reliable detection models. Method. To this end, we present WilDect-YOLO, a deep learning (DL)-based automated high-performance detection model for real-time endangered wildlife detection. In the model, we introduce a residual block in the CSPDarknet53 backbone for strong and discriminating deep spatial features extraction and integrate DenseNet blocks to improve in preserving critical feature information. To enhance receptive field representation, preserve fine-grain localized information, and improve feature fusion, a Spatial Pyramid Pooling (SPP) and modified Path Aggregation Network (PANet) have been implemented that results in superior detection under various challenging environments. Results. Evaluating the model performance in a custom endangered wildlife dataset considering high variability and complex backgrounds, WilDect-YOLO obtains a mean average precision (mAP) value of 96.89%, F1-score of 97.87%, and precision value of 97.18% at a detection rate of 59.20 FPS outperforming current state-of-the-art models. Significance. The present research provides an effective and efficient detection framework addressing the shortcoming of existing DL-based wildlife detection models by providing highly accurate species-level localized bounding box prediction. Current work constitutes a step toward a non-invasive, fully automated animal observation system in real-time in-field applications.     

Increased heavy metal and nutrient contamination does not increase fluctuating asymmetry in the seagrass Halophila ovalis

Fluctuating asymmetry (random differences between symmetric structures, FA) is one of the stress indices used recently to assess a subtle effect of environmental degradation on organisms and is expected to increase under stress conditions. In this study, we developed an original technique of measuring FA in seagrass, Halophila ovalis. We analysed five metric and meristic characters on leaves of the seagrass from a polluted and several control locations in a lagoon in Eastern Australia. The seagrass was sampled from three sites at each location. The analyses revealed significant spatial heterogeneity of samples in fluctuating asymmetry with the highest variability was observed among sites. There was no increase in FA of H. ovalis from polluted location. Possible explanations suggest that whether existing concentrations of heavy metals do not cause developmental stress in seagrass or their effect is compensated or even surpassed by effect of uncontrolled factors.     

Detecting water quality improvement along the Catalan coast (Spain) using stress-specific biochemical seagrass indicators

Evaluating the efficacy of management actions to improve environmental quality is often difficult because there may be considerable lags before ecosystem management actions translate into measurable indicator responses. These delays make it difficult to justify often-expensive remedial actions to prevent eutrophication. Therefore, it is critical to identify reliable, rapid and sensitive indicators to detect degradation and environmental quality improvement. We evaluate the efficacy of a set of indicators based on the seagrass Posidonia oceanica to reliably and quickly detect ecosystem improvements using a 7-year (2003–2010) dataset of 10 stations along the Catalan coast (north-western Mediterranean Sea). In the Catalan region, environmental agencies have invested heavily on wastewater treatment, resulting in significant reductions (ca. 75%) in the BOD₅ discharged to coastal waters from 2003 to 2010. These improvements were clearly reflected at the regional level (i.e. for all the stations averaged) in six biochemical seagrass indicators from our dataset. These indicators were directly related to eutrophication (nitrogen, δ¹⁵N, phosphorus and total non-structural carbohydrates content in rhizomes, δ³⁴S and δ¹³C in seagrass rhizomes and N content in epiphytes). In contrast, seagrass structural indicators, related to seagrass abundance or meadow structure (density, cover) did not show any sign of overall recovery during the monitored period. These results confirm that biochemical seagrass indicators are the most sensitive to water quality improvements within management time-scales (7–10 years) for slow-growing species like P. oceanica. Given the budgetary restrictions under which most management actions operate, the availability of decision-support tools that function at appropriate time-scales is crucial to help managers validate the relative success of their remedial efforts. Our results indicate that low inertia, biochemical seagrass indicators fit this task, and can be a robust set of tools to include in monitoring programmes.     

Halophila ovalis in the Tropical Atlantic Ocean

A species of seagrass in the genus Halophila was found growing in a shallow lagoon on the west shore of Antigua in the Caribbean West Indies. Genetic analysis showed the plants were Halophila ovalis. In addition, the samples had no genetic deviation (using nrDNA sequences) from Halophila johnsonii, considered to be an endemic and endangered species in Florida, USA. Morphological analysis demonstrated the Antiguan Halophila to be well within the range of plant characteristics previously described in the literature for H. ovalis, except for leaf width and number of seeds per fruit, and again, not different from H. johnsonii and very closely related to H. ovalis from the Indo-Pacific. Ours is the first report of H. ovalis in the Tropical Atlantic bioregion.     

U-YOLOv7: A network for underwater organism detection

Detecting and monitoring underwater organisms is very important for sea aquaculture. The human eye struggles to quickly distinguish between aquatic species due to their variety and dense dispersion. In this paper, a deep learning object detection algorithm based on YOLOv7 is used to design a new network, called Underwater-YOLOv7 (U-YOLOv7), for underwater organism detection. This model satisfies the requirements with regards to both speed and accuracy. First, a network combining CrossConv and an efficient squeeze-excitation module is created. This network increases the extraction of channel information while reducing parameters and enhancing the feature fusion of the network. Second, a lightweight Content-Aware ReAssembly of FEatures (CARAFE) operator is used to obtain more semantic information about underwater images before feature fusion. A 3D attention mechanism is incorporated to improve the anti-interference ability of the model in underwater recognition. Finally, a decoupling head using hybrid convolution is designed to accelerate convergence and improve the accuracy of underwater detection. The results show that the network proposed in this paper obtains an improvement of 3.2% in accuracy, 2.3% in recall, and 2.8% in the mean average precision value and runs at up to 179 fps, far outperforming other advanced networks. Moreover, it has a higher estimation accuracy than the YOLOv7 network.     

Recovery and succession in a multi-species tropical seagrass meadow following experimental disturbance: the role of sexual and asexual reproduction

Recolonisation and succession in a multi-species tropical seagrass meadow was examined by creating gaps (50×50 cm) in the meadow and manipulating the supply of sexual and asexual propagules. Measurements of leaf shoot density and estimates of above-ground biomass were conducted monthly to measure recovery of gaps between September 1995 and November 1997. Measurements of the seeds stored in the sediment (seed bank) and horizontal rhizome growth of colonising species were also conducted to determine their role in the recovery process.Asexual colonisation through horizontal rhizome growth from the surrounding meadow was the main mechanism for colonisation of gaps created in the meadow. The seed bank played no role in recolonisation of cleared plots. Total shoot density and above-ground biomass (all species pooled) of cleared plots recovered asexually to the level of the undisturbed controls in 10 and 7 months, respectively. There was some sexual recruitment into cleared plots where asexual colonisation was prevented but seagrass abundance (shoot density and biomass) did not reach the level of unmanipulated controls. Seagrass species did not appear to form seed banks despite some species being capable of producing long-lived seeds.The species composition of cleared plots remained different to the undisturbed controls throughout the 26-month experiment. Syringodium isoetifolium was a rapid asexual coloniser of disturbed plots and remained at higher abundances than in the control treatments for the duration of the study. S. isoetifolium had the fastest horizontal rhizome growth of species asexually colonising cleared plots (6.9 mm day⁻¹). Halophila ovalis was the most successful sexual coloniser but was displaced by asexually colonising species. H. ovalis was the only species observed to produce fruits during the study.Small disturbances in the meadow led to long-term (>2 years) changes in community composition. This study demonstrated that succession in tropical seagrass communities was not a deterministic process. Variations in recovery observed for different tropical seagrass communities highlighted the importance of understanding life history characteristics of species within individual communities to effectively predict their response to disturbance. A reproductive strategy involving clonal growth and production of long-lived, locally dispersed seeds is suggested which may provide an evolutionary advantage to plants growing in tropical environments subject to temporally unpredictable major disturbances such as cyclones.     

Lung Nodule Detection Based on YOLOv3 Deep Learning with Limited Datasets

The early symptom of lung tumor is always appeared as nodule on CT scans, among which 30% to 40% are malignant according to statistics studies. Therefore, early detection and classification of lung nodules are crucial to the treatment of lung cancer. With the increasing prevalence of lung cancer, large amount of CT images waiting for diagnosis are huge burdens to doctors who may missed or false detect abnormalities due to fatigue. Methods: In this study, we propose a novel lung nodule detection method based on YOLOv3 deep learning algorithm with only one preprocessing step is needed. In order to overcome the problem of less training data when starting a new study of Computer Aided Diagnosis (CAD), we firstly pick up a small number of diseased regions to simulate a limited datasets training procedure: 5 nodule patterns are selected and deformed into 110 nodules by random geometric transformation before fusing into 10 normal lung CT images using Poisson image editing. According to the experimental results, the Poisson fusion method achieves a detection rate of about 65.24% for testing 100 new images. Secondly, 419 slices from common database RIDER are used to train and test our YOLOv3 network. The time of lung nodule detection by YOLOv3 is shortened by 2–3 times compared with the mainstream algorithm, with the detection accuracy rate of 95.17%. Finally, the configuration of YOLOv3 is optimized by the learning data sets. The results show that YOLOv3 has the advantages of high speed and high accuracy in lung nodule detection, and it can access a large amount of CT image data within a short time to meet the huge demand of clinical practice. In addition, the use of Poisson image editing algorithms to generate data sets can reduce the need for raw training data and improve the training efficiency.     

Fluorescence imaging application: effect of leaf age on seagrass photokinetics

We used the Imaging-PAM fluorometer to map spatial variability of photosynthesis in three seagrass species, Halophila ovalis, Zostera capricorni and Posidonia australis. Photosynthesis was described by relative photosynthetic rate (PS/50), effective quantum yield (Φ_PSII), non-photochemical quenching (NPQ and qN), electron transport rate (ETR) and leaf absorptivity. Photosynthetic patterns were linked to leaf age and light climate but patterns were not consistent across species. Longitudinal heterogeneity in photosynthesis was apparent along the leaves of all three species while lateral spatial heterogeneity was found only across Z. capricorni and H. ovalis leaves. Age of leaf tissue, determined by longitudinal location on the leaf, strongly influenced photosynthetic activity of Z. capricorni and P. australis. A comparison of H. ovalis leaves of differing maturity demonstrated the influence of leaf age on photosynthetic activity, yet a comparison of Z. capricorni leaves of differing maturity showed no leaf-age effects.Variations in stress-induced changes across a seagrass leaf can be used to identify areas or particular regions of the leaf, which are more susceptible to photodamage. Clear evidence of substantial within-leaf heterogeneity in photosynthetic activity (i.e., a two-fold variation in half saturation constant along a leaf of P. australis) has serious implications for use of small sections of leaf for photosynthetic incubations (such as O₂ or single-point chlorophyll a fluorescence measurements).     

Cutting out the middle clam: lucinid endosymbiotic bacteria are also associated with seagrass roots worldwide

Seagrasses and lucinid bivalves inhabit highly reduced sediments with elevated sulphide concentrations. Lucinids house symbiotic bacteria (Ca. Thiodiazotropha) capable of oxidising sediment sulphide, and their presence in sediments has been proposed to promote seagrass growth by decreasing otherwise phytotoxic sulphide levels. However, vast and productive seagrass meadows are present in ecosystems where lucinids do not occur. Hence, we hypothesised that seagrasses themselves host these sulphur-oxidising Ca. Thiodiazotropha that could aid their survival when lucinids are absent. We analysed newly generated and publicly available 16S rRNA gene sequences from seagrass roots and sediments across 14 seagrass species and 10 countries and found that persistent and colonising seagrasses across the world harbour sulphur-oxidising Ca. Thiodiazotropha, regardless of the presence of lucinids. We used fluorescence in situ hybridisation to visually confirm the presence of Ca. Thiodiazotropha on roots of Halophila ovalis, a colonising seagrass species with wide geographical, water depth range, and sedimentary sulphide concentrations. We provide the first evidence that Ca. Thiodiazotropha are commonly present on seagrass roots, providing another mechanism for seagrasses to alleviate sulphide stress globally.Subject terms: Microbial ecology, Plant ecology, Soil microbiology     

The sterol and fatty acid compositions of seven tropical seagrasses from North Queensland,Australia

The sterol and fatty acid compositions of fresh leaves of the seagrasses Cymodocea serrulata, Enhalus acoroides, Halodule uninervis, Halophila ovalis, H. ovata, H. spinulosa and Thalassia hemprichii are reported. The major fatty acids were palmitic acid, linoleic acid and linolenic acid as expected. H. ovalis and H. ovata were characterized by the relatively high abundance (ca 5%) of the acid hexadeca-7,10,13-trienoic acid (16:3<7 > ). The sterol compositions were typical of higher plants, with sitosterol and stigmasterol accounting for 60–90% of the observed sterols. 28-Isofucosterol was a major sterol (20–30%) only in the Halophila spp. Cluster analysis of the sterol composition data clearly separated the Halophila spp. from the other seagrasses and enabled the distinction of Enhalus sp. from Cymodocea, Halodule and Thalassia spp. The seagrass species were clearly separated into five chemical groups using the combined fatty acid and sterol composition data and the need for a reappraisal of the taxonomic position of Halophila was indicated.     

A case study of utilizing YOLOT based quantitative detection algorithm for marine benthos

Current object detection algorithms suffer from low accuracy and poor robustness when used to detect marine benthos due to the complex environment and low light levels on the seabed. To solve these problems, the YOLOT (You Only Look Once with Transformer) algorithm, a quantitative detection algorithm based on the improved YOLOv4, is proposed for marine benthos in this paper. To improve the feature extraction capability of the neural network, the transformer mechanism is introduced in the backbone feature extraction network and feature fusion part of YOLOv4, which enhances the adaptability of the algorithm to targets in complex undersea environments. On the one hand, the self-attention unit is embedded into CSPDarknet-53, which improves the feature extraction capability of the network. On the other hand, it is transformer-based feature fusion rules that are introduced to enhance the extraction of contextual semantic information in the feature pyramid network. In addition, probabilistic anchor assignment based on Gaussian distribution is introduced to network training. The experimental validation shows that compared with the original YOLOv4, the YOLOT algorithm improves the recognition precision from 75.35% to 84.44% on the marine benthic dataset. The improvement reflects that YOLOT is suitable for the quantitative detection of marine benthos.     

The conservation value of elevation data accuracy and model sophistication in reserve design under sea‐level rise

Many studies have explored the value of using more sophisticated coastal impact models and higher resolution elevation data in sea‐level rise (SLR) adaptation planning. However, we know little about to what extent the improved models and data could actually lead to better conservation outcomes under SLR. This is important to know because high‐resolution data are likely to not be available in some data‐poor coastal areas in the world and running more complicated coastal impact models is relatively time‐consuming, expensive, and requires assistance by qualified experts and technicians. We address this research question in the context of identifying conservation priorities in response to SLR. Specifically, we investigated the conservation value of using more accurate light detection and ranging (Lidar)‐based digital elevation data and process‐based coastal land‐cover change models (Sea Level Affecting Marshes Model, SLAMM) to identify conservation priorities versus simple “bathtub” models based on the relatively coarse National Elevation Dataset (NED) in a coastal region of northeast Florida. We compared conservation outcomes identified by reserve design software (Zonation) using three different model dataset combinations (Bathtub–NED, Bathtub–Lidar, and SLAMM–Lidar). The comparisons show that the conservation priorities are significantly different with different combinations of coastal impact models and elevation dataset inputs. The research suggests that it is valuable to invest in more accurate coastal impact models and elevation datasets in SLR adaptive conservation planning because this model–dataset combination could improve conservation outcomes under SLR. Less accurate coastal impact models, including ones created using coarser Digital Elevation Model (DEM) data can still be useful when better data and models are not available or feasible, but results need to be appropriately assessed and communicated. A future research priority is to investigate how conservation priorities may vary among different SLR scenarios when different combinations of model‐data inputs are used.     

Magnitude of within-patch variation in seagrass Halophila ovalis growth affected by adjacent Thalassia hemprichii vegetation

Seagrass beds in South-east Asia sometimes consist of a mosaic of different species in monospecific patches. We examined whether the magnitude of within-patch variation in the seagrass Halophila ovalis is affected by the presence or absence of surrounding vegetation consisting of another seagrass species Thalassia hemprichii in an intertidal flat in Thailand waters. We measured biomass and growth rates of H. ovalis at the edges and centers of two different types of patches: (i) H. ovalis patches adjoining T. hemprichii vegetation (HT patches), and (ii) H. ovalis patches adjoining unvegetated sand flats (HS patches). Furthermore, we examined the possible effects of interspecific interactions on the growth of H. ovalis by experimentally removing adjoining T. hemprichii at the edges of HT patches. The biomass of H. ovalis was greater at the patch centre than the patch edge in both types of patches. For the growth rate of H. ovalis, significant interactions were detected between patch types and positions in patches. The difference in growth was significant and more than 4-fold between edges and centers of the HS patches, whereas the growth was not significantly different between edges and centers of the HT patches. The removal of T. hemprichii did not significantly affect the growth rate of H. ovalis at the edge of the HT patches. These findings demonstrate that the magnitude of within-patch variation in H. ovalis growth is affected by the conditions of adjoining habitats. However, any effects of local competition with T. hemprichii on H. ovalis growth were not evident in this short-term manipulative experiment.     

MAD-YOLO: A quantitative detection algorithm for dense small-scale marine benthos

Marine biological resources are abundant, and the reasonable development, research and protection of marine biological resources are of great significance to marine ecological health and economic development. At present, underwater object quantitative detection plays a very important role in marine biological science research, marine species richness survey, and rare species conservation. However, the problems of a large amount of noise in the underwater environment, small object scale, dense biological distribution, and occlusion all increase the detection difficulty. In this paper, a detection algorithm MAD-YOLO (Multiscale Feature Extraction and Attention Feature Fusion Reinforced YOLO for Marine Benthos Detection) is proposed, which is based on improved YOLOv5 is proposed to solve the above problems. To improve the adaptability of the network to the underwater environment, VOVDarkNet is designed as the feature extraction backbone. It uses the intermediate features with different receptive fields to reinforce the ability to extract feature. AFC-PAN is proposed as the feature fusion network so that the network can learn correct feature information and location information of objects at various scales, improving the network's ability to perceive small objects. Label assignment strategy SimOTA and decoupled head are introduced to help the model better handles occlusion and dense distribution problems. Experiments show the MAD-YOLO algorithm increases mAP_0.5:0.95 on the URPC2020 dataset from 49.8% to 53.4% compared to the original YOLOv5. Moreover, the advantages of the model are visualized and analyzed by the method of controlling variables in the experimental part. The experiments show that MAD-YOLO is suitable for detecting blurred, dense, and small-scale objects. The model performs well in marine benthos detection tasks and can effectively promote marine life science research and marine engineering implementation. The source code is publicly available at https://github.com/JoeNan1/MAD-YOLO.