Cotton Boll Growth Status Recognition Method under Complex Background Based on Semantic Segmentation

Abstract

In order to quickly and accurately identify the state of cotton bells during different growth periods in complex backgrounds between cotton fields, this paper studied cotton and cotton bells under natural light, uses machine positions from different angles. We compare 4 classes of semantic segmentation models such as PSPnet, FCN, deeplabv3+ and SegNet and propose an optimization algorithm for the structure of PSPnet models. Morphologies, size, and weed background positions were segmented for cotton bells and cotton, and cotton growth periods were classified and identified according to the state and type of segmentation. Based on the pspnet model, more and accurate cotton bell feature information was extracted in a complex background by using the context information of different regions as a prior knowledge and selecting ResNet50 as the backbone feature extraction network. At the same time, the improved model coding part incorporates the context coding module, complements more global prior knowledge, and integrates the cotton shallow features in the model decoding part, solving the problem of insufficient segmentation accuracy under the influence of the complex background. After training using 1400 images, 400 field cotton bell images were used as tests. The results show that the improved model outperformed the original model in accuracy and speed, the identification method that uses the segmentation results to judge the cotton bell growth stage is satisfied.

Abstract from the paper’s OpenAlex record, shown for indexing and research. Copyright remains with the authors or the publisher. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.