An underwater, multispectral light source

Group members: Bhrugurajsinh Pradyumansinh Chudasama, Candice Murray, Anirban Chatterjee

Introduction

The objective of this project is to create an underwater multispectral light source that can fit in a Go-Pro housing. Underwater multispectral imaging has varied uses, from monitoring health of coral reefs to estimating plankton density in marine waters. A multispectral light source also helps in characterization and measurement of absorption of different wavelengths in water, which can provide estimates of pollution levels in water. By enclosing the entire system inside a Go-Pro housing, we not only ensure that the system is water-proof but also extremely portable. Designing the system without application-specific integrated circuits also means that the system can be programmed and modified to adjust brightness of the light sources as well as alter the pattern in which the light sources are flashed.

Background

Motivation

Multispectral imaging

Multispectral imaging is a method of finding the reflectance of surfaces by analyzing images of the surfaces. When the illumination levels and wavelengths are well known, the reflectances can be found based on pixel values of an image using the relation

$m_{j}=g_{j}\Delta \sum _{\lambda =1}^{N}s_{j,\lambda }r_{\lambda }I_{\lambda }$

where $m_{j}$ is the pixel value of the jth pixel, $g_{j}$ is the pixel scaling, $s_{j,\lambda }$ is the camera sensitivity, $r_{\lambda }$ is the reflectance and $I_{\lambda }$ is the intensity.[1]

The reflectance values can provide information about the state of the environment being imaged. Many organisms, such as coral, exhibit a characteristic reflectance spectrum. There is also often a change in reflectance when organisms are healthy versus unhealthy. Some examples of this are [2] and [3].

Challenges of Underwater Imaging

Absorption of light in water

Water exhibits much higher absorption than air does at some wavelengths of light. This can be seen in the graphs below.

Increasing the depth of the water increases the absorption, which can be modeled by the absorption equation $I(x)=I(x_{o})*exp(-\alpha (x-x_{o}))$ where $\alpha$ is the absorption coefficient for water at the wavelength of interest and $I(x)$ and $I(x_{o})$ are the intensities at the final and starting locations, respectively.

Scattering in water

Due to particulate matter in the water, backscattering can have a large impact on underwater imaging. Backscattering occurs when the illumination used to take a picture reflects off of particles close to the camera, and can result in spotty images or images where the visibility distance is very short. For examples of images with backscattering, see [4].

The most common method around this is to move the illumination source far away from the camera detector so that any scattering that occurs won't reflect into the camera sensor.

Methods

LED Design

Our system needs to drive 7 high brightness LEDs. We selected 6 Philips LUMILEDS series LEDs to provide illumination in the visible wavelengths. The wavelengths of these LEDs have been chosen such that they are evenly spread out in the visible spectrum. The 7th LED was a ultra-violet (UV) LED (365 nm) to be used primarily for exciting fluorescence in plankton and other underwater organisms.

LED specifications from the manufacturer are shown in the table below. For full LED specifications, see [5].

LED Color	Red-Orange	PC Amber	Lime	Green	Cyan	Blue
Lumens @ 350mA	72 lm	78 lm	167 lm	102 lm	76 lm	41 lm
Lumens @ 700mA	134 lm	140 lm	313 lm	161 lm	122 lm	70 lm
Efficacy @ 350mA	98 Lm/W	73 Lm/W	174 lm/W	100 Lm/W	75 Lm/W	38 Lm/W
Efficacy @ 700mA	83 Lm/W	63 Lm/W	160 lm/W	68 Lm/W	51 Lm/W	29 Lm/W
Typical Wavelength	617 nm	591 nm	567 nm	530 nm	505 nm	470 nm
Wavelength Range	610 to 620 nm	588 to 592 nm	566 to 569 nm	520 to 540 nm	490 to 515 nm	460 to 485 nm
Beam Angle	125°	120°	125°	125°	125°	125°
Recommended Operating Current	700 mA	350 mA	700 mA	700 mA	700 mA	700 mA
Maximum Rated Drive Current	700 mA	700 mA	1000 mA	1000 mA	1000 mA	1000 mA
Typical Forward Voltage	2.1 Vf	3.05 Vf	2.75 Vf	2.9 Vf	2.9 Vf	2.95 Vf
Maximum Forward Voltage	2.8 Vf	3.51 Vf	3 Vf	3.51 Vf	3.51 Vf	3.51 Vf
Thermal Resistance	8 C°/W	10.4 C°/W	6.4 C°/W	10.4 C°/W	10.4 C°/W	10.4 C°/W
Max Recommended Junction Temp	135 °C	130 °C	150 °C	150 °C	150 °C	150 °C
Operating Temperature Range	-40 to 120 °C	-40 to 110 °C	-40 to 135 °C	-40 to 135 °C	-40 to 135 °C	-40 to 135 °C
Dimensions L x W x H	4.5 x 3 x 2 mm	10 x 10 x 3.7 mm	10 x 10 x 3.7 mm	10 x 10 x 3.7 mm	10 x 10 x 3.7 mm	10 x 10 x 3.7 mm

The flux as a function of distance traveling through the water for these LEDs is shown in the graph below. For these calculations the absorption coefficients are those of pure water for the central wavelength of each LED. Solid lines represent the LED when run at maximum current, dashed lines are running at half maximum current.

To address the underwater challenges outlined in the background section, the high brightness LEDs were chosen to combat absorption, and the capability to adjust the brightness level was designed into the circuit. This allowed us to adjust the LEDs so that wavelengths with higher absorption could be driven harder than LEDs with less absorption so that the illuminance detected at the camera could be relatively similar for different wavelengths. This also allows for accounting for the varying flux output levels of the different LEDs.

To combat backscattering, the LEDs that would require higher radiance (those with longer wavelengths and thus higher absorption) were moved further away from the camera sensor to minimize the backscattering. Unfortunately within the very constrained dimensions of the underwater housing for the camera we weren't able to do more physically to combat backscattering. However, there are many image processing tricks that could be used to reduce or eliminate backscattering, such as taking two images at slightly different angles and removing the spots in the image that move the most, as these will most likely be backscattering effects.

Circuit Design

The high current requirement for these LEDs(~700mA) means that we need to use a driver circuit to drive these LEDs as micro controllers cannot sink/source more than 25mA of current. Since the brightness of these LEDs should be adjustable, we need to have some form of LED dimming capability incorporated in our system.

This brightness control was done through pulse width modulation. The system included an Atmega168 micro controller, which has six hardware PWM channels as shown in the diagram below:

The project requirement was to drive seven LEDs. Hence one of the general purpose IO pins, pin # 6, on the controller was programmer to emulate the hardware PWM output via software programming. The switch on the outside of the GoPro casing is used to provide interrupt to the controller. This leads to hardware debouncing issue, where the mechanical switch creates multiple glitches on the interrupt pin input. Following circuit was used to remove the glitches:

Apart from that the software debouncing was implemented in the code. When the controller receives the interrupt, it enters halts the program execution and jumps to the interrupt service routine (ISR). In the ISR, a small delay was added before servicing the interrupt.

On every interrupt the controller switches to the next state. The program on the controller has eight states, seven states for enabling the corresponding PWM channels, and one states for switching off all the PWM channels. The PWM outputs drive the gate of the above mentioned NMOS transistors.

Following is the link to the GITHUB repo for the code:

GITHUB repo - Multispectral-PWM-lighting

Our design uses an NMOS transistor to sink about 0.7A though a high brightness LED. By controlling the gate voltage of an NMOS device, we can control the current flowing through the LED and hence the brightness of the LED. We chose the ZVN4306A FET from Diodes Incorporated as our high current FETs. From hspice simulations, we found that, varying the gate voltage from ~1.7V to 3.3V led to a current sweep of 0.15A t 0.70A through the LEDs. This gave us ballpark estimates of the gate voltages we should be using. Since the Atmega168 cannot generate analog signals, we fed the output of the PWM pins to the gate via a resistor. This leads to low-passing the PWM signals; effectively generating an analog voltage at the gate of the FET. This is shown in the following diagrams:

Once we were satisfied with the design and had tested the system on a breadboard, we designed a 2 layer PCB for the controller in PCBExpress. Snapshots of the PCB are shown below:

Snapshots of the parts of the assembly are shown below:

The final assembly after sealing into the waterproof housing:

Results

The system was tested after completion by capturing images of the Macbeth color checker. The experimental setup is shown below.

The tank was filled with water 15 cm deep, and the Macbeth color checker was held down using rocks. Rulers were placed along the edges of the tank to ensure that the camera returned to the same position each time an image was taken (it was necessary to move the camera each time because switching the LEDs caused the camera to go into a menu, which it was necessary to look at the front-facing screen to get out of and return to image capture mode.) The camera was placed at the top of the water, with the LEDs and lens just submerged. All of the images were taken in a dark room. Experiments were run with about 350 mA drive current in an effort to preserve the battery lifetime.

The resulting images are shown below for two different runs of the experiment.

In the first run there was more inconsistency in the camera location because the ruler guides weren't used. In the second experimental run, ruler guides were used to help with consistency, but luminance levels were lower due to the battery beginning to run out.

Conclusions

An LED system for multispectral imaging was designed, built and tested successfully. This design is currently in prototype stage of development, and there is still much to be done. Some potential directions of future development of this system are:

1. Making the system more robust.

2. Testing the performance of our system with LEDs at higher brightness setting.

3. Using higher power density batteries.

4. Using a lensing system to focus the light.

5. Modifying the switch to a method that doesn't also switch things on the GoPro camera.

6. Synchronizing the GoPro shutter release with the LEDs.

7. 3D printing a new underwater housing that would be less space constrained.

References

1. J. Breneman, H. Blasinski, J. Farrell, "The color of water: Using underwater photography to estimate water quality," Proc. SPIE 9023, Digital Photography X, 90230R (7 March 2014) [6]

2. E.J. Hoochberg, M.J. Atkinson and S. Andrefouet, "Spectral reflectance of coral reef bottom-types worldwide and implications for coral reef remote sensing," Remote Sensing of Environment, 85, 159-173, (2003) [7]

3. K.E. Joyce and S.R.Phinn, "Hyperspectral analysis of chlorophyll content and photosynthetic capacity of coral reef substrates," Limnology and Oceanographyy, 48, 489-496, (2003). [8]

4. H. Buiteveld and J. M. H. Hakvoort and M. Donze, "The optical properties of pure water," in SPIE Proceedings on Ocean Optics XII, edited by J. S. Jaffe, 2258, 174--183, (1994). [9]

5. K. S. Shifrin, Physical Optics of Ocean Water, American Institute of Physics, New York, (1988). [10]

6. "Optical Absorption of Water Compendium", [11]

7. M. Parmar, S. Lansel, J. Farrell, "An LED-based lighting system for acquiring multispectral scenes," [12]

Appendix

For this project, Anirban Chatterjee was the analog circuit lead, Bhrugurajsinh Chudasama was the digital circuit lead and Candice Murray was the optical lead. Anirban designed the circuit board used to control the LEDs, Bhrugurajsinh programmed the microcontroller that determined the switching of the LEDs, and both did general circuit testing and debugging and designing of the debouncing system. Candice did the designs for the LEDs and testing of completed system and worked on features necessary to fit the system within the footprint of the underwater GoPro housing.