This dataset tracks 17 patent families published between 2015 and mid-2026 that combine phenotyping platform claims with machine learning, computer vision or deep learning methods. That is a small, specialised corpus compared to broader agtech categories, reflecting how narrowly the search string targets the AI-plus-phenotyping intersection rather than agricultural imaging in general. Because publication lags filing by around 18 months, the true 2025-2026 filing count is almost certainly higher than what has published so far.

No single company holds a dominant position in this corpus; CROCUS LABS GMBH holds the most-cited representative filing, covering a controlled-environment cultivation system with an integrated machine learning module, while much of the remaining activity comes from individual researchers and academic co-filing groups rather than large corporate portfolios. The co-assignee network shows every pairing at a single joint filing, which is the pattern of an early-stage, not-yet-consolidated field. Anyone scanning for an acquisition target or licensing partner should look at filing quality and citation counts rather than portfolio size, since none of the tracked assignees have built a large one yet.

Filings rose from 2 in 2017 to a peak of 9 in 2022 and have declined since, with the most recent tracked years showing no filings at all. Because 2022 sits at the midpoint of the observed range, the honest read is flat-to-declining rather than growing, though the last one to two years are undercounted due to publication lag. If you are timing an entry, treat the current data as a lull rather than confirmed contraction until another 12-18 months of publications land.

The most-cited record in this dataset, US20240096092A1 and its WO counterpart, both centre on automated hyperspectral vegetation index derivation for high-throughput phenotyping — a claim area that a new entrant working with hyperspectral imaging should review closely. Separately, US20220400620A1 claims a controlled-environment cultivation system combining gallium-nitride LED arrays, an imaging device and a machine learning aggregator module, which is narrower and easier to design around by changing the light source, sensor configuration or ML architecture. Neither of these blocks pure field-based, non-controlled-environment phenotyping approaches, which sit outside their claim scope.

Relative to the core imaging and AI classes (G06K, G06T, G06V, G06N), sub-areas like hyperspectral index calibration, UAV-based multispectral capture, below-ground root phenotyping and edge-deployed field inference show thinner direct coverage in this corpus. That does not guarantee freedom to operate, since a 17-family dataset can miss adjacent filings outside this exact search string, but it does suggest these branches carry a lower density of competing claims. A first filing that ties a specific sensor modality to a specific trait-extraction model in one of these branches has more room to establish a clean claim than one written around general computer-vision trait detection.