The invention provides a Chaetoceros image segmentation method based on unsupervised pixel-by-pixel classification. The method comprises the following steps: extracting seta information of an originalimage I0 by using a GSDAM image processing algorithm to obtain a seta information image G, and extracting cell edges of the original image I0 by using a Canny image processing algorithm to obtain a cell edge image C; generating an automatic training sample by using the seta information image G and the cell edge image C, and selecting a positive sample and a negative sample; inputting the positivesample and the negative sample into a deep convolutional neural network DCNN of a DeepLab for pixel-by-pixel training; inputting unknown labeled pixels in the original image I0 into the trained model, dividing an unknown labeled pixel region into a Chaetoceros target and a background, and obtaining a final segmentation result image. The segmentation method provided by the invention can automatically segment Chaetoceros cells from an original microscopic image, has a good segmentation effect, and provides a good research basis for subsequent identification and classification of Chaetoceros.