Single-frame autofocusing using multi-LED illumination
An imaging method is provided that includes: (i) providing a microscope having a lens and an autofocusing camera positioned adjacent to the microscope; (ii) positioning an illumination source adjacent to the microscope; (iii) moving a sample to a predefined offset position and illuminating the sample; (iv) acquiring an image of the illuminated sample via the autofocusing camera; and (v) utilizing a convolution neural network to identify an in-focus position of the sample. The convolution neural network may further include an input layer, output layer, and at least one hidden layer situated between the input and output layers. The hidden layer(s) may be selected from a group consisting of a convolution layer, pooling layer, normalization layer, fully connected layer, and a combination thereof. The convolution neural network may be trained to accurately define the weight to be applied to the layer(s). The illumination source may be a single-LED, dual-LED, LED array, Köhler illumination, and a combination thereof. The convolution neural network may advantageously predict the in-focus position of the acquired image without axial scanning and may be a multi-domain convolution neural network, which may receive input(s) selected from spatial features, Fourier transform of the acquired image, autocorrelation of the acquired image.
1. An imaging method, comprising:
providing a microscope having a lens and an autofocusing camera positioned adjacent to the microscope;
positioning a first and second LEDs adjacent to the microscope;
moving a sample to a predefined offset position and illuminating the sample from two different incident angles via the first and second LEDs;
acquiring an image of the illuminated sample via the autofocusing camera, the image including two image copies of the sample; and
utilizing a convolution neural network to identify an in-focus position of the sample.
2. The imaging method of claim 1 , wherein the convolution neural network further comprises an input layer and output layer.
3. The imaging method of claim 2 , wherein the convolution neural network further comprises at least one hidden layer situated between the input and output layers; and wherein the hidden layers are selected from a group consisting of a convolution layer, pooling layer, normalization layer, fully connected layer, and a combination thereof.
4. The imaging method of claim 3 , wherein the convolution neural network is trained to accurately define the weight applied to the layers.
5. An imaging method, comprising:
providing a microscope having a lens and an autofocusing camera positioned adjacent to the microscope;
positioning an illumination source adjacent to the microscope;
moving a sample to a predefined offset position and illuminating the sample;
acquiring an image of the illuminated sample via the autofocusing camera; and
utilizing a convolution neural network to identify an in-focus position of the sample.
6. An imaging method of claim 5 , wherein the convolution neural network further comprises an input layer and output layer.
7. The imaging method of claim 5 , wherein the convolution neural network further comprises at least one hidden layer situated between the input and output layers; and wherein the hidden layers are selected from a group consisting of a convolution layer, pooling layer, normalization layer, fully connected layer, and a combination thereof.
8. The imaging method of claim 5 , wherein the convolution neural network is trained to accurately define the weight applied to the layers.
9. The imaging method of claim 5 , wherein the illumination source is selected from a group consisting of a single-LED, dual-LED, LED array, Köhler illumination, and a combination thereof.
10. The imaging method of claim 5 , wherein the illumination occurs beneath the sample.
11. The imaging method of claim 5 , wherein the illumination is relayed to a back focal plane of a condenser lens.
12. The imaging method of claim 5 , wherein the illumination is created by the microscope.
13. The imaging method of claim 5 , wherein the convolution neural network predicts the in-focus position of the acquired image without axial scanning.
14. The imaging method of claim 5 , wherein the convolution neural network is a multi-domain convolution neural network, wherein at least one input for the multi-domain convolution neural network is selected from a group consisting of spatial features, Fourier transform of the acquired image, autocorrelation of the acquired image, discrete cosine transform of the acquired image, discrete wavelet transform of the acquired image, and a combination thereof.