TasselNetV2plus

TasselNetV2plus performs high-throughput plant counting from high-resolution RGB imagery to support agricultural phenotyping and yield-related analyses.

Key Features:

• Object counting paradigm: Directly regresses plant counts from imagery rather than relying on bounding-box object detection.
• Efficiency and speed: Implements minor modifications to TasselNetV2 to achieve approximately an order-of-magnitude speedup, reaching ~30 frames per second on 1980 × 1080 images.
• Accuracy retention: Maintains the same level of counting accuracy as the original TasselNetV2 despite reduced computational demand.
• Crop validation: Validated on wheat ears counting, maize tassels counting, and sorghum heads counting.
• Comparison to object detectors: Addresses computational challenges of conventional object detectors such as Faster R-CNN when applied to high-resolution images.

Scientific Applications:

• Plant phenotyping workflows: Provides accurate count data for stages including seed breeding, germination, cultivation, fertilization, and pollination.
• Yield estimation and harvesting: Supplies high-resolution count estimates useful for yield estimation and harvest planning.
• UAV and high-throughput platforms: Suited for integration with unmanned aerial vehicle (UAV) imagery and other high-throughput phenotyping platforms.

Methodology:

Profiled computational bottlenecks in the original TasselNetV2 and implemented minor, strategic modifications while retaining the direct count-regression object counting paradigm.