Research on Apple Image Segmentation in Natural Environment Based on Deep Learning

. To accurately identify apples from the complex background in the natural environment and help the apple harvesting robot harvest apples accurately, an improved apple image segmentation algorithm based on Deeplabv3 framework is proposed, which is named as AppleDNet. Using the famed Deeplabv3 algorithm, combined with Atrous convolution, Depthwise separable convolution and transfer learning, not only can achieve more accurate segmentation results but also improve segmentation speed. In addition, the traditional image filtering algorithm is adopted to obtain a smoother segmentation image and effectively eliminate the image stitching trace. The experimental results demonstrate that the performance of the proposed method is superior to the original Deeplabv3 and other popular mainstream image segmentation algorithms, with an overall accuracy of 97.90%, which has certain significance and advantages in practice.

Paper

Full text

PDF

Research on Apple Image Segmentation in Natural Environment Based on Deep Learning

Semantic Scholar · Computer Science · 2021

Abstract

. To accurately identify apples from the complex background in the natural environment and help the apple harvesting robot harvest apples accurately, an improved apple image segmentation algorithm based on Deeplabv3 framework is proposed, which is named as AppleDNet. Using the famed Deeplabv3 algorithm, combined with Atrous convolution, Depthwise separable convolution and transfer learning, not only can achieve more accurate segmentation results but also improve segmentation speed. In addition, the traditional image filtering algorithm is adopted to obtain a smoother segmentation image and effectively eliminate the image stitching trace. The experimental results demonstrate that the performance of the proposed method is superior to the original Deeplabv3 and other popular mainstream image segmentation algorithms, with an overall accuracy of 97.90%, which has certain significance and advantages in practice.

Similar papers

© 2026 NYSGPT2525 LLC