研究目的
Investigating the development of a robust segmentation algorithm for kiwifruit detection under varying illumination and complex background using a double-layer pulse-coupled neural network (PCNN) model.
研究成果
The proposed double-layer PCNN model accurately segments ripe kiwifruit targets under varying illumination and complex background, achieving an average misclassification rate of 4.75%. Future work includes optimizing the algorithm and extending its application to more challenging fruit targets.
研究不足
Presence of holes and small incorrect segmentation blocks on segmented kiwifruit objects due to non-smooth surfaces and dark spots. The parameter λ in the fusion step affects segmentation quality and is determined experimentally.
1：Experimental Design and Method Selection:
The study employs a double-layer PCNN model for image segmentation, integrating frequency-tuned saliency and total variation model for feature fusion.
2：Sample Selection and Data Sources:
Over 100 test images captured under varying lighting conditions by a digital camera with 320 × 240 pixels.
3：List of Experimental Equipment and Materials:
Digital camera, TMS320DM642 DSP with 4M × 64bit SDRAM.
4：Experimental Procedures and Operational Workflow:
Pre-segmentation in the first layer of PCNN to determine optimal color-difference, feature fusion using total variation model, and re-segmentation in the second layer of PCNN.
5：Data Analysis Methods:
Subjective visual evaluation and objective quantitative assessment using misclassification error (ME) metric.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
