In order to realize the image segmentation for tomato plants at night, an improved pulse coupled neural network (PCNN) image segmentation algorithm was designed based on the maximum inter-group variance method. The algorithm weighted the link input in the traditional PCNN model. Before the image segmentation, the threshold was obtained based on the maximum inter-class variance (Otsu) algorithm, and then the threshold was assigned to the weight of the link input, the synaptic link coefficient , the link weight amplification factor and the threshold iterative decay time constant in the improved PCNN model. The results of 849 images of tomato plants at night showed that the average segmentation accuracy was 90.43% and the average segmentation time of one image was 0.9944s. The weighted processing of the link input could reduce the number of the iterations of improved PCNN and improve the real-time performance of the algorithm. Based on the Otsu algorithm, the network parameters can be set adaptively in the improved PCNN model. The comparative analysis based on the visual evaluation, the maximum entropy and the segmentation accuracy rate showed that the segmentation effect of improved PCNN model was better than those of the Otsu algorithm and the traditional PCNN model, and its real-time performance was also better than that of the traditional PCNN model.