Traditionally, therefore, accurate and efficient detection and counting of rice panicles are labor intensive. Based on this, this paper introduces RiceNet, a CNN that achieves high performance of detecting and counting rice panicles from high-resolution images. RiceNet exploits advanced deep learning techniques that achieve better accuracy and speed than the conventional ones. RiceNet has a compact convolutional layer-based architecture to extract features efficiently that also incorporates attention layers to capture high-order dependencies, hence making the exact detection under varying lighting and occlusion conditions. The time complexity of the model is made small enough to efficiently analyze large image data sets with high throughput. RiceNet achieves high accuracy and computational efficiency over traditional image processing and other CNN architectures on diverse rice field images of different rice varieties and stages of growth. Notably, the model can yield timely estimates of crop yield and manages the crop within 30 seconds, which is a significant reduction in panicle detection time. The future work will optimize RiceNet for broader application on more cereal crops and wider agricultural applications to further liberate its potential to revolutionize precision farming.
Kushwaha et al. (Fri,) studied this question.