MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202641084731 A) filed by Vidyavardhaka College Of Engineering on July 10, 2026, for Deep Learning Model For Weed Classification In Sorghum Plant.
Inventors include Dr. Naveen Kumar H N; Prof. Anusha R; Samiksha R; Sanjana Srinivasa; and Prof. Meghana S.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The present invention discloses an intelligent edge-enabled autonomous weed classification and selective weed removal system for sorghum cultivation using deep learning and embedded robotic automation. The invention comprises a lightweight convolutional neural network based on MobileNetV2 trained through transfer learning using field images of sorghum plants, grass weeds, and broadleaf weeds. The trained model is optimized using TensorFlow Lite quantization and deployed on an embedded Raspberry Pi platform to enable low-latency inference with minimal computational resources. A vision acquisition module continuously captures real-time crop images, which are preprocessed and supplied to the deep learning classifier for discrimination between crop plants and weed species. Upon identification of a weed, a decision control module activates a motor driver, relay circuitry, and a mechanically actuated cutting mechanism for localized weed removal while preventing damage to sorghum plants. An ultrasonic sensing module enables obstacle detection and safe navigation during field operation, whereas display and alert modules provide real-time operational feedback. The integrated architecture performs autonomous image acquisition, embedded inference, navigation, weed localization, selective mechanical actuation, and continuous field monitoring without dependence on cloud computing or herbicide application. The invention substantially reduces manual labour, minimizes indiscriminate herbicide usage, decreases operational costs, and improves precision weed management through real-time edge intelligence. Owing to its lightweight computational requirements, scalability, and adaptability to precision agriculture platforms, the proposed system is suitable for deployment in small- and large-scale agricultural environments and can be extended to other crop ecosystems with minimal retraining.
Disclaimer: Curated by HT Syndication.
