Automatic Weed Detection using CCOA based YOLO Network in Soybean Field

Due mostly to competition with crops, weeds now account for $45 \%$ of crop failures in the gardening sector. This ratio can be decreased via an actual wild plant detecting technique. Because of the large variety of weed species that are mixed together and the complicated distribution status of soybean, traditional approaches have a poor weed detection rate. To successfully apply selective herbicide spraying, weeds in crop fields must be accurately identified. The opaque allocation of wildflowers and cuts prepares it difficult to identify weeds because border segmentation at the overlap is imprecise, making it impossible to accurately classify pixels. This paper suggests a perfect for weed finding in soybean fields based on an enhanced YOLOv5 network architecture to address this issue. A new weed identification model is produced as a consequence of fusing feature maps of various sizes in the decoding step in order to improve the model’s utilisation of global information linkages. The CrissCross Optimisation (CCO) algorithm chooses the best kernel for convolution. The average ratio reached $98 \%$ when this model was used to identify a dataset comprising many widely dispersed weedy soybean seedlings, and the accuracy increased by $3.94 \%$ in comparison to other traditional semantic segmentation models. The findings provided guidance for the additional use of modifiers in wildflower credit by demonstrating that the model suggested in this paper can positively speech the subjects of improper border delineation credit when wildflowers are thickly dispersed with harvests and improper organisation when credit boards overlay.