A highly sensitive underwater video system for use in turbid aquaculture ponds

The turbid, low-light waters characteristic of aquaculture ponds have made it difficult or impossible for previous video cameras to provide clear imagery of the ponds’ benthic habitat. We developed a highly sensitive, underwater video system (UVS) for this particular application and tested it in shrimp ponds having turbidities typical of those in southern Taiwan. The system’s high-quality video stream and images, together with its camera capacity (up to nine cameras), permit in situ observations of shrimp feeding behavior, shrimp size and internal anatomy, and organic matter residues on pond sediments. The UVS can operate continuously and be focused remotely, a convenience to shrimp farmers. The observations possible with the UVS provide aquaculturists with information critical to provision of feed with minimal waste; determining whether the accumulation of organic-matter residues dictates exchange of pond water; and management decisions concerning shrimp health.


Results and Discussion
Monitoring shrimp feeding in turbid water. In most countries of Southeast Asia, including Taiwan and China, shrimp farms usually grow animals at high densities (> 100 shrimps m −2 ) in closed greenwater ponds or biofloc systems. The water used, therefore, often is highly turbid due to suspended material, phytoplankton, or both. Usually, visibility is approximately 100 cm, but sometimes less than 50 cm (Fig. 1). Mean values of total suspended material (TSM) in the three shrimp ponds (P1, P2, P3) on the NSYSU campus ranged between 4.4 and 48 mg/L, consistent with the range in lakes and commercial shrimp ponds across southern Taiwan (3.8 to 50 mg/L, Fig. 2). Turbidity in the UVS images presented here, therefore, is representative of conditions in Taiwanese shrimp farms.
The low-light intensity characteristic of turbid water has proven problematic for camera systems in the absence of auxiliary lighting 8 . The UVS, however, functions well even at 15 lux or less. Selected UVS images (snapshot mode) serve to demonstrate the system's ability to visualize evidence of shrimp feeding, even in aquaculture  P-1 to P-3 (within red rectangle) represent concentrations of TSM in shrimp ponds 1, 2, and 3 respectively. Other TSM concentrations are from lakes (labeled L-A and L-B) or shrimp/fish farms (labeled P-A1 to P-G; PA-1 and PA-2 represent duplicate sampling at pond A) in southern Taiwan. ponds. Figure 3 contains images of shrimp ponds 1, 2, and 3 before (panels A,C,E) and after (panels B,D,F) provision with formulated feed pellets, indicated by red circles. In low-turbidity ponds 1 and 2 (panels A-D), shrimp are visible before and during feeding. Even in pond 3, characterized by high TSM and bioflocs, feeding pellets are visible on the sediment surface (panels E,F).
Traditionally, shrimp farmers determine the rate and extent of food consumption using a feeding tray to which pellets are added. The tray is lowered into the pond, retrieved after some tens of minutes, and inspected to infer shrimp feeding. This method relies heavily on the farmer's experience and can be misrepresentative in cases of feeding lag or over feeding during the short time the tray is deployed. In contrast, clear, real-time images of food pellets and shrimp feeding behavior provided by the UVS allow a temporally integrated fine-tuning of food supply. Less well resolved images taken in highly turbid waters (e.g., Pond 3) also provide important information, unavailable from feeding-tray assessments, permitting farmers to adjust the amount of feed accordingly. Overall, therefore, the UVS largely improves on the traditional feeding-tray method to monitor shrimp feeding.
Acoustic control systems, used to improve feeding productivity in fish cages and shrimp farms 5,6 , are suitable for clear-water systems or shrimp ponds of large size (e.g., > 10 ha). Acoustic control may not be cost effective in Taiwan, Thailand, or China, however, because most shrimp farms in those countries range in size from approximately 0.16 to several ha 2 . In addition, a high percentage of intensive shrimp farms in Taiwan and Thailand are owned by small-scale operators 2 who unlike their large-scale counterparts, do not require or cannot afford pioneering instrumentation. Although the UVS is not without cost, it provides not only real-time feeding information, but additional insights as well (next section).
Evaluating excess organic matter, shrimp size, and shrimp health. Jackson et al. 3 reported that sediments in shrimp ponds contain 14% (in terms of total nitrogen) of formulated food delivered to the shrimp. This excess organic matter, plus that from shrimp feces, phytodetritus, etc., provides a substantial nutrient source for bacteria, some of which may be pathogenic. To reduce bacterial load overall and keep water quality high in Thailand, shrimp ponds are heavily aerated to concentrate particulate organic waste at their center, rather than have it distributed randomly across the pond bottom 2 . Confirming a concentrated distribution, however, has previously not been easily accomplished. The UVS makes possible rapid detection of waste-rich organic sediment in turbid waters or greenwater systems, for example, the thick layer of organic detritus and formulated feed pellets seen in Fig. 4A,B. With clear imagery of the benthic surface, shrimp farmers can decide when and to what degree to clean pond sediments or exchange fouled pond water with fresh seawater.
The UVS also can be used to estimate shrimp size in situ, precluding the need to collect them with nets for length measurements. For example, with reference to a ruler attached to a brick, we estimated two shrimp in Fig. 4B each to be approximately 10 cm long. In a calibration exercise, these estimates correlated strongly with manual measurements of shrimp length: y = 1.04x-0.4, r = 0.99, p < 0.005, n = 20, where y is estimated size (cm) and x = measured size (cm) (Fig. 5). Deployment of a standard-sized mesh (e.g., from netting) on the sediment surface would facilitate size estimates of multiple shrimp and account for changing perspective with distance from the camera.
One can assess the health status of shrimp according to their swimming and feeding behaviors and body transparency 12    by poor water quality or disease, including lethargy, sluggish swimming, and spiral swimming, can be easily seen. Given its ability to monitor shrimp feeding, behavior, and internal organs, even in turbid waters, the UVS can serve farmers as an early warning system to detect unhealthy shrimp, making possible an immediate response to overfeeding, disease, or poor water quality.
Providing real-time underwater images to aquaculture farmers. Asia is the biggest aquaculture production area in the world and 70% of global aquaculture products are produced by China and Taiwan 13,14 . In January 2016, a lethal cold snap in those two regions killed aquaculture stock including fish, clams, and shrimp 15 . In Taiwan alone, losses were estimated at more than $100M US 16 . Fish farmers were able to promptly harvest some types of cold-stunned fish, e.g., milkfish and tilapia, and reduced their economic loss. Losses of demersal fish and shrimp were profound, however, in large part because they could not be seen in high-turbidity water. In particular, 2-to 3-year old groupers weighing 10 to 30 kg were killed but unseen until they floated to the surface of fish ponds several days after the extremely cold weather. They were spoiled and unharvestable. In cases such as these rare cold snaps (ca. once per decade), the UVS could supply fish farmers with images pertinent to making harvest decisions to save their stocks or at least to reduce their property loss (Fig. 4C). Struthers  Is there an effect on shrimp behavior at very low light levels (< 15 lux) when the IR bulbs are activated? Some fish are sensitive to IR light in highly turbid habitats 18 , but shrimp likely are not, based on reports of their spectral sensitivity. Vision in the penaeid shrimp, Sergestes similis, has a single peak of sensitivity from 480 to 520 nm 19 (Lindsay et al. 19 ). Frank and Widder 20 tested vision in 12 species of mesopelagic crustaceans, including shrimp, and found all had a strong peak within the range 470 to 500 nm. In investigating the effects of different light sources on growth of Litopenaeus vannamei, You et al. 21 showed the shrimp grew rapidly when illuminated by a metal halide lamp (MHL). The spectrum of a typical MHL, however, is between 385 nm and 674 nm, suggesting only visible light affects shrimp growth. Finally, Sanudin et al. 22 concluded that lighting conditions, including IR wavelengths, did not affect feeding by post larvae (PL10 and older) of Penaeus vannamei. Therefore, although we did not run IR sensitivity experiments, the literature suggests shrimp are insensitive to IR light.

Advantages of the UVS and its future improvements.
Judicious siting of the video cameras, which currently "see" only a fixed area, can allow assessment of the pond overall. To increase coverage for large shrimp ponds, farmers could install additional cameras; one server can control nine cameras based on current technology. In the future, we envision rotating UVS, even mobile ones, to increase the area monitored. Such systems could be connected with software to monitor and control the ponds' environment based on real-time imagery. That is, the UVS is a first step in developing fully automated systems to detect the amount of food available to the shrimp, provide food when needed, assess the degree of accumulation of sediment in the ponds, and even differentiate between low light levels caused by turbidity vs absence of sunlight. We understand conceptually how machine learning, integrated with the UVS, can achieve these goals and hope in the future to investigate them.

Materials and Methods
Underwater video system (UVS). The underwater video system (UVS) contains a 500-GB hard drive (expandable up to 2-TB), a server, and a power supply (each camera operates at 12 V and 500 mA) supporting three custom manufactured underwater cameras; the system can be configured with as many as nine cameras (Fig. 1). Each camera is emplaced autonomously. The real-time image from each is processed by a built-in digital signal processor (DSP) and sent to a server through a power over Ethernet (PoE) cable, then transported to an internet account via a wireless network.
Each camera has two 3-watt infrared (780 nm) LED bulbs, contained within a waterproof, aluminum alloy (Aluminum 6061-T6) housing with an acrylic window. Cameras were fixed approximately 20~30 cm above the bottom of shrimp ponds to monitor shrimp activity, feeding status, welfare, and benthic sediments. The UVS captures color images when the light intensity is higher than 15 lux; below that level, the infrared LED bulbs are controlled by an IR-Cut sensor and turn on automatically, generating black and white images. The system is reliable in seawater shrimp ponds (salinity ranging from 30 to 34); the UVS operated at depths of 1.2 to 1.8 m for more than 12 months. The acrylic window needs to be cleaned of biofouling every 3 to 4 days. On sunny and most cloudy days, the UVS operates in color mode because the light intensity at the bottom of the ponds is approximately 790 to 1,580 lux (about 1% of surface light intensity, based on annual sunlight intensity data in southern Taiwan (www.cwb.gov.tw). Therefore, the UVS can be used in low transparency waters during the daytime without extra light sources. The underwater images can be accessed remotely by a smart phone or through the internet.
Video camera specifications. The camera (OV2710) is a true full HP (1080p) CMOS image sensor designed specifically to deliver high-end HD video to a digital video camcorder with a display resolution of 1920 × 1080 pixels, operating at 30 frames per sec (for more details, see the OV2710 commercial website, OV2710 product brief). Built with Omni Vision's proprietary 3 μ m OmniPixeI3-HS high sensitivity pixel technology, the Scientific RepoRts | 6:31810 | DOI: 10.1038/srep31810 OV2710 delivers low light sensitivity of 3300 mV/lux-sec, S/N ratio of 39 dB, dark current of 10 mV/sec, and a peak dynamic range of 69 dB, which in combination allow the camera to operate in conditions ranging from strong light to nearly complete darkness (below 15 lux), such as in highly turbid aquaculture ponds, even at night, so that 24 hour surveillance of shrimp or fish is possible (Fig. 4D).

Measurement of total suspended matter (TSM). Concentrations of TSM in lakes and aquaculture
ponds in southern Taiwan were measured using the method of Hung et al. 23 . Briefly, 20 to 150 ml of water was filtered through a tared 25 mm polycarbonate filter (pore size = 0.4 μ m), then the filter was gently rinsed with 15 to 20 mL of Milli-Q water to remove salts. The filter was dried, weighed on a micro balance, and the mass of suspended matter was calculated by difference.
Application of UVS in aquaculture ponds of different turbidity. We conducted observations in three shrimp ponds on the campus of National Sun Yat-Sen University from June 2014 to November 2015. Each pond contained between 40 to 70 tonnes of seawater and supported at a minimum several thousand white shrimp, Penaeus vannamei. Shrimp feed pellets were evenly dispensed three times daily. TSM concentrations in ponds 1, 2, and 3 ranged from 4.6 to 6.4, 5.2 to 8.2, and 27 to 73 mg/L, respectively (Fig. 2).
We observed shrimp behavior and feeding via real-time UVS. The UVS image stream was recorded, allowing review of previous activities (see video in Supplemental Materials). We evaluated benthic surface conditions with respect to feed pellets, fecal pellets, and excess organic matter. We estimated the length of shrimp that swam or crawled adjacent to a ruler (15 cm long) affixed to a brick. We calibrated our length estimates using shrimp individually measured by hand, placed in a large tank filled with shrimp pond water, and measured again using the UVS and a ruler.